Multi-modal large model domain knowledge database migration method

By classifying and optimizing the transmission characteristics of data in the multimodal large model database, the problem of unstable data transmission rate in database migration is solved, fast and stable migration efficiency is achieved, and the data correction process is simplified.

CN118885461AInactive Publication Date: 2024-11-01ZHIZI ENGINE (BEIJING) INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410972630.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-01
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art does not set a set of good transmission rules during database migration, resulting in unstable data migration rate and affecting migration efficiency.

Method used

By initially sorting and classifying data in different formats in the main database, determining the pending partition, and determining the transmission correlation interval based on the data transmission characteristics, optimizing the data transmission rate; verifying the migrated single data, identifying and processing abnormal periods, Ensure consistency of data capacity.

Benefits of technology

It realizes stability and fast transmission rate during database migration, improves data migration efficiency, simplifies the data correction process, and reduces correction time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118885461A_ABST
    Figure CN118885461A_ABST
Patent Text Reader

Abstract

The invention discloses a domain knowledge database migration method of a multi-modal large model, relates to the technical field of database migration, and solves the problems that a group of better transmission rules is not set and the migration rate of data is gradually optimized. Sorting the single data in the same type of partitions, identifying a corresponding transmission association interval from the transmission data characteristics of the previous group of single data, taking the transmission association interval as an execution interval of the next group of single data, determining a related transmission association interval in the transmission process of the next group of single data, and storing the related transmission association interval in the next group of single data; and by parity of reasoning, the transmission association interval is optimized step by step, so that the whole database is stable enough in the migration process, the transmission rate of the whole database is high enough, and a better data migration effect is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of database migration, and in particular to a method for migrating domain knowledge databases of multimodal large models. Background Art

[0002] Large multimodal models need to use knowledge bases in the fields of visual understanding, visual generation, and unified visual models. Large multimodal models are an important development direction in the field of artificial intelligence in recent years. The core of these models is to integrate information from different modalities, such as text, images, audio, and video, through deep learning models to achieve cross-modal understanding and generation. In order to build effective large multimodal models, it is necessary to widely absorb and integrate knowledge and technologies from multiple fields.

[0003] The application with publication number CN117971801A discloses a domestic big data migration method and system based on new communications, which relates to the field of big data migration technology. In the present invention, the user determines the database to be migrated out, the migration data range and the database to be migrated in. The system will analyze and determine the features of the migration-out database and the features of the migration-out database. The first necessary migration data list in the large model database is determined by the migration-out database features. The remaining data is screened according to the migration data range with the help of the knowledge graph of the core network network element to determine the second migration data list that meets the subdivided field. Finally, according to the features between the libraries and the migration strategy, the data of the database to be migrated out that meets the first migration data list and the second migration data list is migrated to the database to be migrated in, so as to achieve the advantages of flexibility and scalability of the new generation of communication technology, and realize the migration of the database of the large model to the database of different subdivided fields, thereby greatly promoting the development of artificial intelligence.

[0004] During the migration process, the database generally transfers the relevant data of the database to another database based on the set transmission rules. However, the original migration process will affect the migration rate of the data due to the conversion between data formats. In addition, a set of better transmission rules are not set during the data migration process to gradually optimize the data migration rate in order to ensure the stability and timeliness of the data during the migration process. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for migrating domain knowledge databases of multimodal large models, which solves the problem of not setting a set of better transmission rules to gradually optimize the migration rate of data.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for migrating domain knowledge database of a multimodal large model, comprising the following steps:

[0007] S1) Determine the main database used by this large model and the auxiliary database to be migrated, sort out the data in different formats in the main database, classify the different single data contained in the main database, and confirm the partitions to be processed. The specific sub-steps are:

[0008] S11, confirming the format of the single data stored in different folders in the main database. If there are multiple sets of single data in different formats, the single data with the same format are divided into the same partition to be processed, and the single data with different capacities in the same partition to be processed are sorted in ascending order;

[0009] S12. If the plurality of groups of single data stored in the main database are in the same format, the plurality of groups of single data are divided into the same partition to be processed, and the single data of different capacities in the same partition to be processed are sorted in ascending order;

[0010] S2), based on the determined partition to be processed, data migration is performed on the single data in the partition to be processed, based on the data transmission characteristics of the corresponding single data during the data migration process, the transmission association interval is determined from the data transmission characteristics, and the transmission association interval in the previous group of single data transmission process is used as the execution interval in the next group of single data transmission process, the specific sub-steps are:

[0011] S21, migrate the single data sorted at the first position in the partition to be processed, and the migration direction is from the primary database to the auxiliary database, and synchronously monitor the data transmission rate of the primary database and the read and write rate of the auxiliary database during the data migration process, and generate a data transmission rate change curve and a read and write rate change curve according to different data transmission rates and different read and write rates corresponding to different moments, and both sets of change curves are located in the same two-dimensional coordinate system;

[0012] S22, determining the fluctuation points in the transmission rate change curve, the change trends of the points before and after the fluctuation points are opposite, for each group of fluctuation points determined, constructing a group of correlation lines that pass through the fluctuation points and are perpendicular to the X-axis, and determining the corresponding correlation lines for each group of fluctuation points, and the correlation lines all intersect with the data transmission rate change curve and the read / write rate change curve;

[0013] S23, confirm the related areas between adjacent related lines, the upper and lower side lines of the related areas are partial line segments of the data transmission rate change curve and the read / write rate change curve, and the left and right side lines are partial line segments of the corresponding related lines. Based on the confirmed groups of related areas, the coverage area FM of the corresponding related areas is confirmed. i , where i represents different correlation regions and determines the time length J between the correlation lines before and after this correlation region i , using T i =FMi ÷J i Confirm the relevant features of its relevant area T i ;

[0014] S24. Different related features T confirmed from different related regions i In the example, select the minimum value T i min, mark the relevant area as the standard area, determine the several groups of data transmission rates associated with the standard area before and after, and determine its transmission association interval based on the minimum and maximum values ​​of its data transmission rates;

[0015] S25, when the single data in the partition to be processed that is located in the group after the first single data is transmitted, this transmission association interval is executed, and the data of the single data in the transmission process is analyzed, and the transmission association interval is reconfirmed and used as the execution interval in the process of the next group of single data transmission, and so on, until the transmission of several single data in the partition to be processed is completed;

[0016] S26: For other partitions to be processed in the master database, the same processing method as steps S21-S25 is adopted to transmit a plurality of single data;

[0017] S3) Perform data verification for each group of single data that has completed the transmission, identify the capacity change of the single data before and after the migration, and based on the identification result, determine whether such single data has migration abnormalities and calibrate the abnormal single data. The specific method is as follows:

[0018] S31, determine the data capacity of this single data in the main database and mark it as R k , where k represents different single data, and then the data capacity of this single data in the auxiliary database is determined and marked as L k ;

[0019] S32, if R k ≠L k , which means that there is a migration anomaly in the migration process of this single data, and this single data is marked as abnormal single data. If R k =L k , indicating that there is no abnormality in the migration process of this single data;

[0020] S4), for the abnormal single data with calibration, identify the transmission data characteristics of such abnormal single data during the migration process, and identify the abnormal period therefrom, confirm the migration data associated with the abnormal period, confirm the capacity value of the confirmed migration data, identify whether the capacity difference of the migration data is consistent with the capacity difference of the abnormal single data, and generate a migration error signal based on the identification result; the specific method is:

[0021] S41, based on the transmission data characteristics during the migration process, determining a data transmission rate change curve and a read / write rate change curve of the abnormal single data, and both sets of change curves are located in the same two-dimensional coordinate system;

[0022] S42, identifying the data change trend of the same period from the two groups of change curves, if the data change trends of the two groups are the same, no processing is performed, if the data change trends of the two groups are different, marking the same period as an abnormal period;

[0023] S43, based on the determined abnormal period, determine the migration data migrated in the abnormal period, and determine whether the capacity data of the migration data in the primary database and the secondary database are the same. If they are the same, no processing is performed. If they are different, the capacity difference Cz is determined, and Cz>0. The capacity differences of different migration data determined in different abnormal periods are summed to determine the total capacity difference ZZ;

[0024] S44, identify the data capacity R of the abnormal single data in the main database k and the data capacity L located in the auxiliary database k , determine the associated difference Gz of the two data capacities, where Gz = |R k -L k |, if Gz≠ZZ, generate a migration error signal of abnormal single data, and synchronously display the migration data with the determined capacity difference; if Gz=ZZ, display the migration data with the determined capacity difference.

[0025] The present invention provides a method for migrating domain knowledge database of a multimodal large model. Compared with the prior art, it has the following beneficial effects:

[0026] The present invention divides the data of the same format in the migration process into the same partitions, sorts the single data in the same partitions, identifies the corresponding transmission association interval from the transmission data characteristics of the previous group of single data, uses the transmission association interval as the execution interval of the next group of single data, determines the relevant transmission association interval from the transmission process of the next group of single data, and uses it as the execution interval of the next group of single data again, and so on, and gradually optimizes the transmission association interval, so that the entire database is not only sufficiently stable during the migration process, but also has a fast transmission rate, so as to achieve a better data migration effect;

[0027] The migrated single data is verified to identify whether it is abnormal, and the transmission data of the abnormal single data is analyzed and confirmed to identify the abnormal period. The migration data related to the abnormal period is confirmed, and the capacity change of the migrated data is analyzed first to determine whether the capacity change is consistent. This method is more convenient for external operators to correct data, without the need for operators to search the entire single data, thus ensuring the correction efficiency during the data correction process and fully reducing the correction time. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the present invention;

[0029] Figure 2 Schematic diagram for determining the relevant areas of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Example 1

[0032] Multimodal large models need to use knowledge bases in the fields of visual understanding, visual generation, and unified visual models. Multimodal large models are an important development direction in the field of artificial intelligence in recent years. The core of multimodal large models is to integrate information from different modalities, such as text, images, audio, and video, through deep learning models.

[0033] See also Figure 1 The present application provides a method for migrating domain knowledge database of a multimodal large model, comprising the following steps:

[0034] S1) Determine the main database used by this large model and the auxiliary database to be migrated, and initially organize the data of different formats in the main database, so that the different single data contained in the main database are classified, and the partitions to be processed are confirmed. Specifically, the main database and the auxiliary database are determined in advance by the operator. After the relevant determination is completed, the main database includes several groups of related data in different formats, and the related data are all single data. The single data can be understood as the related data stored in different folders in the corresponding main database, and in order to extract data corresponding to the multimodal large model, the format of the related data stored in each different folder is the same. Among them, the specific sub-steps of determining the partitions to be processed are:

[0035] S11, confirming the format of the single data stored in different folders in the main database. If there are multiple sets of single data in different formats, the single data with the same format are divided into the same partition to be processed, and the single data with different capacities in the same partition to be processed are sorted in ascending order;

[0036] S12. If the plurality of groups of single data stored in the main database are in the same format, the plurality of groups of single data are divided into the same partition to be processed, and the single data of different capacities in the same partition to be processed are sorted in ascending order;

[0037] Specifically, before the original folder of the main database is sorted, the position mark related to the single data inside it does not change. Later, after all the groups of single data are transferred to the auxiliary database, they are re-sorted according to the original internal position mark, so that the sorting method in the auxiliary database is consistent with the sorting method stored in the main database. Since the corresponding folders will have relevant marks after the sorting process is completed, they will not be described in detail here, and this application mainly focuses on the rate during the database migration process and the processing analysis of abnormal data;

[0038] S2), based on the determined partition to be processed, data migration is performed on the single data in the partition to be processed, based on the data transmission characteristics of the corresponding single data during the data migration process, the transmission association interval is determined from the data transmission characteristics, and the transmission association interval in the previous group of single data transmission processes is used as the execution interval in the next group of single data transmission processes. Specifically, in order to make different single data more stable and faster during the transmission process, the transmission process characteristics of the corresponding data can be used, and the transmission characteristics include transmission rate and read and write rate. The read and write rate is the reading and writing speed of the auxiliary database for the migrated data. Since the same format data has transmission characteristics when being transmitted, the best transmission mode can be gradually determined from the transmission characteristics, and gradually optimized, so that the entire database is not only sufficiently stable during the migration process, but also has a fast enough transmission rate;

[0039] The specific sub-steps of determining the transmission association interval are as follows:

[0040] S21, migrate the single data sorted in the first position in the partition to be processed, and the migration direction is from the primary database to the auxiliary database, and synchronously monitor the data transmission rate of the primary database and the read and write rate of the auxiliary database during the data migration process, and generate the data transmission rate change curve and the read and write rate change curve according to the different data transmission rates and different read and write rates corresponding to different time periods, and the two sets of change curves are both located in the same two-dimensional coordinate system, the horizontal coordinate axis of the two-dimensional coordinate system is the timeline, and the vertical coordinate axis is the rate parameter;

[0041] S22, determine the fluctuation point in the transmission rate change curve, the change trends of the points before and after the fluctuation point are opposite (that is, when the value from the front end point to the fluctuation point is in a climbing state, then after reaching this fluctuation point, its value will be in a decreasing state. Similarly, when the value from the front end point to the fluctuation point is in a decreasing state, then the value from this fluctuation point to the next set of points will be in an increasing state), for each group of fluctuation points determined, construct a group of correlation lines that pass through this fluctuation point and are perpendicular to the X-axis, determine the corresponding correlation lines for each group of fluctuation points, and the correlation lines intersect with the data transmission rate change curve and the read-write rate change curve;

[0042] S23, Combination Figure 2 , confirm the related areas between adjacent related lines, the upper and lower side lines of the related areas are partial line segments of the data transmission rate change curve and the read / write rate change curve, and the left and right side lines are partial line segments of the corresponding related lines. Based on the confirmed groups of related areas, confirm the coverage area FM of the corresponding related areas i , where i represents different correlation regions and determines the time length J between the correlation lines before and after this correlation region i , using T i =FM i ÷J i Confirm the relevant features of its relevant area T i ;

[0043] S24. Different related features T confirmed from different related regions i In the above example, we select a set of minimum values ​​T i min, mark the relevant area as the standard area, determine the several groups of data transmission rates associated with the standard area before and after, and determine its transmission association interval based on the minimum and maximum values ​​of its data transmission rates;

[0044] Specifically, after the relevant area is confirmed, the smaller the area in the corresponding area and the longer the associated time, the smaller the area of ​​the corresponding area per unit time, which means that the transmission rate and the read / write rate of the corresponding area are closest. In the process of data migration, when the two groups of rates are closest, the data migration process is the most stable and the migration effect is the best. Therefore, by confirming the rates in such areas, the data migration effect of the next group of data in the same format can be guaranteed. By performing gradual optimization, the best data migration effect can be achieved.

[0045] S25, when the single data in the partition to be processed that is located in the group after the first single data is transmitted, this transmission association interval is executed, and the data of the single data in the transmission process is analyzed, and the transmission association interval is reconfirmed and used as the execution interval in the process of the next group of single data transmission, and so on, until the transmission of several single data in the partition to be processed is completed;

[0046] S26: For other partitions to be processed in the master database, the same processing method as steps S21-S25 is adopted to transmit a number of single data;

[0047] Specifically, if there are five groups of single data in the partition to be processed, namely A, B, C, D and E, where A is the single data in the first position, a group of transmission-related intervals will be confirmed during data transmission. This transmission-related interval is determined from the data transmission process of A and used as the execution interval during the data transmission process of B, that is, the transmission rate of the B data is limited during transmission. When the B data is transmitted, another group of transmission-related intervals can be confirmed and used as the execution interval of the C data, and so on. By step-by-step confirmation, the transmission rate can be gradually reduced to ensure the stability and rate of the data during the transmission process.

[0048] Example 2

[0049] In the specific implementation process, compared with the above-mentioned embodiment, this embodiment mainly focuses on the verification of migration data, identifying whether there is abnormal capacity of the migration data, and avoiding data loss of such migration data;

[0050] S3) For each group of single data that has completed the transmission, data verification is performed to identify the capacity change of the single data before and after the migration, and based on the identification result, it is determined whether such single data has migration abnormalities and the abnormal single data is calibrated, wherein the specific method of determination is:

[0051] S31, determine the data capacity of this single data in the main database and mark it as R k , where k represents different single data, and then the data capacity of this single data in the auxiliary database is determined and marked as L k ;

[0052] S32, if R k =L k , indicating that there is no abnormality in the migration process of this single data. If R k ≠L k , which means that there is a migration anomaly in the migration process of this single data, and this single data is marked as abnormal single data;

[0053] S4), for the abnormal single data with calibration, identify the transmission data characteristics of such abnormal single data during the migration process, and identify the abnormal period therefrom, confirm the migration data associated with the abnormal period, confirm the capacity value of the confirmed migration data, identify whether the capacity difference of the migration data is consistent with the capacity difference of the abnormal single data, and generate a migration error signal based on the identification result, wherein the specific method of identification is:

[0054] S41, based on the transmission data characteristics during the migration process, determining a data transmission rate change curve and a read / write rate change curve of the abnormal single data, and both sets of change curves are located in the same two-dimensional coordinate system;

[0055] S42, identifying the data change trend of the same period from the two groups of change curves. If the data change trends of the two groups are the same, no processing is performed. If the data change trends of the two groups are different, the same period is marked as an abnormal period. For example, the time period in the curve is 1-2-3-4-5-6-7-8-9. In the same period of 1-2, the trends of the data in the two curves are both climbing. If the data change trend of one curve is climbing and the data change trend of the other curve is falling, it means that the data change trends of the two groups are different. The same period can be determined from the different segments of the data change trend, and the corresponding period belongs to the abnormal period.

[0056] S43, based on the determined abnormal period, determine the migration data migrated during the abnormal period, and determine whether the capacity data of the migration data in the primary database and the secondary database are the same. If they are the same, no processing is performed. If they are different, determine the capacity difference Cz, and Cz>0 (that is, the maximum value minus the minimum value of the two sets of capacity data), and sum the capacity differences of different migration data determined in different abnormal periods to determine the total capacity difference value ZZ;

[0057] S44, identify the data capacity R of the abnormal single data in the main database k and the data capacity L located in the auxiliary database k , determine the associated difference Gz of the two data capacities, where Gz = |R k -L k |, if Gz=ZZ, the migration data with the determined capacity difference will be displayed for external related operators to correct the data. If Gz≠ZZ, a migration error signal of the abnormal single data will be generated, and the migration data with the determined capacity difference will be displayed synchronously. For such migration error signals, external related operators will need to correct the abnormal single data, give priority to correcting the data errors in the migration data, and then compare and calibrate other data within the abnormal single data, and make relevant adjustments to ensure that the single data structure after migration is complete and there is no data missing.

[0058] Specifically, by analyzing the migration process of the corresponding data, the abnormal time period with abnormal migration process can be identified based on the corresponding migration characteristics. Based on the determined abnormal time period, the associated migration data is searched, and the capacity change of the migration data is analyzed first to determine whether its capacity change is consistent. This method is more convenient for external relevant operators to correct data, and there is no need for relevant operators to perform relevant searches on the entire single data, thereby ensuring the correction efficiency in the data correction process and substantially reducing the correction time.

[0059] Example 3

[0060] The specific implementation process of this embodiment includes the entire implementation process of the above two groups of embodiments.

[0061] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0062] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for migrating domain knowledge database of a multimodal large model, characterized in that: The following steps are involved: S1) Determine the main database used by this large model and the auxiliary database to be migrated, sort out the data in different formats in the main database, classify the different single data contained in the main database, and confirm the partitions to be processed; S2), based on the determined partition to be processed, data migration is performed on the single data in the partition to be processed, based on the data transmission characteristics of the corresponding single data during the data migration process, a transmission association interval is determined from the data transmission characteristics, and the transmission association interval in the previous group of single data transmission processes is used as the execution interval in the next group of single data transmission processes; S3) Perform data verification for each group of single data that has completed the transmission, identify the capacity change of the single data before and after the migration, and based on the identification result, determine whether such single data has migration abnormality and calibrate the abnormal single data; S4) For the calibrated abnormal single data, identify the transmission data characteristics of such abnormal single data during the migration process, and identify the abnormal time period therefrom, confirm the migration data associated with the abnormal time period, confirm the capacity value of the confirmed migration data, identify whether the capacity difference of the migration data is consistent with the capacity difference of the abnormal single data, and generate a migration error signal based on the identification result.

2. The method for migrating domain knowledge database of a multimodal large model according to claim 1, characterized in that: In step S1, the specific sub-steps of determining the partition to be processed are: S11, confirming the format of the single data stored in different folders in the main database. If there are multiple sets of single data in different formats, the single data with the same format are divided into the same partition to be processed, and the single data with different capacities in the same partition to be processed are sorted in ascending order; S12: If the several groups of single data stored in the main database are all in the same format, the several groups of single data are divided into the same partition to be processed, and the single data of different capacities in the same partition to be processed are sorted from small to large.

3. The method for migrating domain knowledge database of a multimodal large model according to claim 2 is characterized in that: In step S2, the specific sub-steps of determining the transmission association interval are: S21, migrate the single data sorted at the first position in the partition to be processed, and the migration direction is from the primary database to the auxiliary database, and synchronously monitor the data transmission rate of the primary database and the read and write rate of the auxiliary database during the data migration process, and generate a data transmission rate change curve and a read and write rate change curve according to different data transmission rates and different read and write rates corresponding to different moments, and both sets of change curves are located in the same two-dimensional coordinate system; S22, determining the fluctuation points in the transmission rate change curve, the change trends of the points before and after the fluctuation points are opposite, for each group of fluctuation points determined, constructing a group of correlation lines that pass through the fluctuation points and are perpendicular to the X-axis, and determining the corresponding correlation lines for each group of fluctuation points, and the correlation lines all intersect with the data transmission rate change curve and the read / write rate change curve; S23, confirm the related areas between adjacent related lines, the upper and lower side lines of the related areas are partial line segments of the data transmission rate change curve and the read / write rate change curve, and the left and right side lines are partial line segments of the corresponding related lines. Based on the confirmed groups of related areas, the coverage area FM of the corresponding related areas is confirmed. i , where i represents different correlation regions and determines the time length J between the correlation lines before and after this correlation region i , using T i =FM i ÷J i Confirm the relevant features of its relevant area T i ; S24. Different related features T confirmed from different related regions i In the example, select the minimum value T i min, mark the relevant area as the standard area, determine the several groups of data transmission rates associated with the standard area before and after, and determine its transmission association interval based on the minimum and maximum values ​​of its data transmission rates; S25. When the single data in the partition to be processed that is located behind the first single data is transmitted, this transmission association interval is executed, and the data of the single data during the transmission process is analyzed, and the transmission association interval is reconfirmed and used as the execution interval during the transmission of the next group of single data, and so on, until the transmission of several single data in the partition to be processed is completed.

4. The method for migrating domain knowledge database of a multimodal large model according to claim 3 is characterized in that: The step S2 also includes: S26. For other partitions to be processed within the main database, the same processing method as steps S21-S25 is adopted to transmit several single data.

5. The method for migrating domain knowledge database of a multimodal large model according to claim 1, characterized in that: In step S3, the specific method for determining abnormal single data is: S31, determine the data capacity of this single data in the main database and mark it as R k , where k represents different single data, and then the data capacity of this single data in the auxiliary database is determined and marked as L k ; S32, if R k ≠L k , indicating that there is a migration anomaly in the migration process of this single data, and this single data is marked as abnormal single data.

6. The method for migrating domain knowledge database of a multimodal large model according to claim 5, characterized in that: In step S32, if R k =L k , indicating that there is no abnormality in the migration process of this single data.

7. The method for migrating domain knowledge database of a multimodal large model according to claim 1, characterized in that: In step S4, the specific method of identifying whether the capacity difference of the migration data is consistent with the capacity difference of the abnormal single data is: S41, based on the transmission data characteristics during the migration process, determining a data transmission rate change curve and a read / write rate change curve of the abnormal single data, and both sets of change curves are located in the same two-dimensional coordinate system; S42, identifying the data change trend of the same period from the two groups of change curves, if the data change trends of the two groups are the same, no processing is performed, if the data change trends of the two groups are different, marking the same period as an abnormal period; S43, based on the determined abnormal period, determine the migration data migrated in the abnormal period, and determine whether the capacity data of the migration data in the primary database and the secondary database are the same. If they are the same, no processing is performed. If they are different, the capacity difference Cz is determined, and Cz>0. The capacity differences of different migration data determined in different abnormal periods are summed to determine the total capacity difference ZZ; S44, identify the data capacity R of the abnormal single data in the main database k and the data capacity L located in the auxiliary database k , determine the associated difference Gz of the two data capacities, where Gz = |R k -L k |, if Gz≠ZZ, a migration error signal of abnormal single data is generated, and the migration data with the determined capacity difference is displayed synchronously.

8. The method for migrating domain knowledge database of a multimodal large model according to claim 7, characterized in that: In the step S44, if Gz=ZZ, the migration data of the determined capacity difference is displayed.