Cloud platform standard compliance evaluation method and system based on large language model

Abstract

The invention belongs to the technical field of large language models, and particularly relates to a cloud platform standard compliance evaluation method and system based on a large language model. Storing the standard document of the cloud platform to a distributed object storage database; constructing a secondary index based on elastic search, and performing full-text retrieval and management on the standard document; adjusting the large language model combination based on the standard document; using the adjusted large language model combination to form test content; based on the task scheduling engine and the test content, the evaluation task is allocated to the corresponding test object for testing, and result data is obtained; and receiving result data, generating a test report and task progress information in combination with the adjusted large language model combination, and outputting the test report and the task progress information. The method has the advantages that whole-process informatization management can be achieved, evaluation tasks are distributed to corresponding test objects for testing based on a large language model, a test report and task progress are output, the standardized evaluation requirement of cloud computing is met, and whole-chain traceability is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of large language models, and in particular to a cloud platform standard compliance evaluation method and system. Background Art

[0002] The cloud computing standard evaluation system mainly solves the problem of API (Application Programming Interface) compliance testing, and uses low-code technology to realize the visual arrangement of interfaces in cloud computing industry standards, which simplifies the evaluation process and improves efficiency. The system as a whole includes platform management, task management, result management and report management modules. Platform management is responsible for authentication access information processing, task management manages the evaluation task process, result management displays and preliminarily analyzes the evaluation results, and report management is used to generate test reports in a fixed format. Each module has a different functional focus, forming a system with a functional foundation and a relatively fixed evaluation scope.

[0003] The existing cloud computing standard evaluation system lacks intelligent language understanding and analysis technology for documents, resulting in fixed test content and test logic, high change costs, and cannot adapt to the dynamic development of cloud technology. The evaluation system is limited to the compliance evaluation of a specific type of standard interface, and implements functions one by one based on the understanding of R&D personnel. It cannot adapt to the evaluation tasks of large hybrid cloud architecture systems, and cannot perform test management for functional items without interfaces. Due to the frequent updates of standards in the cloud computing field, the existing evaluation system cannot be dynamically updated according to changes in standards, and it is difficult to meet the standardized evaluation needs of the new generation of cloud computing. Summary of the invention

[0004] The purpose of the present invention is to provide a cloud platform standard compliance evaluation method based on a large language model to solve the above technical problems;

[0005] The purpose of the present invention is also to provide a cloud platform standard compliance evaluation system based on a large language model to solve the above technical problems;

[0006] The cloud platform standard compliance evaluation method based on the large language model includes:

[0007] Step S1, storing the standard document of the cloud platform in a distributed object storage database;

[0008] Step S2, constructing a secondary index based on elastic search to perform full-text retrieval and management on the standard document;

[0009] Step S3, adjusting the large language model assembly based on the standard document to obtain the adjusted large language model assembly;

[0010] Step S4, formulating test content using the adjusted large language model combination;

[0011] Step S5, based on the task scheduling engine and the test content, assign the evaluation task to the corresponding test object for testing to obtain result data;

[0012] Step S6, receiving the result data, generating a test report and task progress information in combination with the adjusted large language model assembly, and then outputting the resultant test report and task progress information.

[0013] Preferably, the secondary index in step S2 performs millisecond-level full-text retrieval on the standard document.

[0014] Preferably, the large language model assembly in step S3 includes a Chinese vector model and a Qwen2.5 base large model.

[0015] Preferably, step S6 comprises,

[0016] Step S61, receiving and generating the test report according to the result data and the report template;

[0017] Step S62, determining the task progress information according to the result data;

[0018] Step S63, after the test report is approved by a preset approval strategy, the task progress information and the approved test report are released.

[0019] Preferably, in step S6, the test content, the result data and the approval strategy are visualized through a low-code engine and a large-screen design engine.

[0020] Cloud platform standard compliance evaluation system based on large language model, including:

[0021] A storage and retrieval module, used for storing standard documents and performing full-text retrieval and management of the standard documents;

[0022] A model auxiliary module is connected to the storage and retrieval module, obtains the standard document, and adjusts the large language model assembly based on the standard document to obtain the adjusted large language model assembly;

[0023] A subject module, connected to the model auxiliary module, formulates test content according to the adjusted large language model combination;

[0024] Platform module, used to record basic information of the object under test;

[0025] A task module, connected to the platform module and the subject module, assigns assessment tasks based on the test content and basic information of the tested object, and then performs the test to obtain result data;

[0026] The result module is connected to the task module and the model auxiliary module, receives the result data, generates a test report and task progress information in combination with the adjusted large language model assembly, and then outputs it.

[0027] Preferably, the storage and retrieval module comprises:

[0028] A storage unit, storing the standard document via a distributed object storage database;

[0029] The retrieval unit is connected to the storage unit, constructs a secondary index based on elastic search, and performs full-text retrieval and management on the standard document.

[0030] Preferably, the test content includes set cloud computing subjects, data processing subjects, metadata subjects, network subjects and newly created test subjects.

[0031] Preferably, the platform module enters the basic information of the object under test through low code.

[0032] Preferably, the result module includes:

[0033] A data receiving unit, connected to the task module, for receiving the result data;

[0034] A data combining unit, connected to the data receiving unit, for checking the result data and obtaining the checked result data;

[0035] A report generating unit, connected to the data combining unit, generates the test report for the verified result data according to a preset report template and the adjusted large language model combination;

[0036] A progress confirmation unit, connected to the data combination unit, for determining the task progress information according to the verified result data;

[0037] A report approval unit, connected to the report generation unit, for approving the test report according to a preset approval strategy;

[0038] A report publishing unit is connected to the progress confirmation unit and the report approval unit, and is used to publish the task progress information and the approved test report.

[0039] The beneficial effect of the present invention is that the above technical scheme can realize the information management of the whole process, and based on the large language model, the evaluation tasks can be assigned to the corresponding test objects for testing, and the test reports and task progress can be output to meet the standardized evaluation requirements of cloud computing and realize full-chain traceability. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1It is a step diagram of the cloud platform standard compliance evaluation method based on a large language model of the present invention;

[0041] Figure 2 is a schematic diagram of step S6 of the present invention;

[0042] Figure 3 is a schematic diagram of a cloud platform standard compliance evaluation system based on a large language model of the present invention;

[0043] Figure 4 is a connection block diagram of the storage and retrieval module of the present invention;

[0044] Figure 5 It is a connection block diagram of the result module of the present invention;

[0045] Figure 6 It is a technical architecture diagram of the cloud platform standard compliance evaluation system based on a large language model of the present invention;

[0046] Figure 7 It is a logic diagram of the cloud platform standard compliance evaluation system based on a large language model of the present invention;

[0047] Figure 8 is a schematic diagram of an embodiment of the present invention.

[0048] In the attached figure: 1. Storage and retrieval module; 11. Storage unit; 12. Retrieval unit; 13. Distributed object storage database; 14. Secondary index; 15. Qwen2.5 base large model; 16. Low code engine; 17. Chinese vector model; 18. Large screen design engine; 19. Task scheduling engine; 2. Model auxiliary module; 3. Subject module; 31. Cloud computing subject; 32. Data processing subject; 33. Metadata subject; 34. Network subject; 4. Platform module; 41. X platform; 42. Y platform; 43. Z platform; 5. Task module; 6. Result module; 61. Data receiving unit; 62. Data compound unit; 63. Report generation unit; 64. Progress confirmation unit; 65. Report approval unit; 66. Report publishing unit; 7. Evaluation task; 8. Report template; 9. Test report; 10. Standard compliance detection large screen. DETAILED DESCRIPTION

[0049] 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.

[0050] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0051] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0052] Cloud platform standard compliance evaluation method based on large language model, such as Figure 1 , Figure 6 As shown, including,

[0053] Step S1, storing the standard document of the cloud platform into the distributed object storage database 13;

[0054] Step S2, constructing a secondary index 14 based on elastic search to perform full-text retrieval and management on standard documents;

[0055] Step S3, adjusting the large language model assembly based on the standard document to obtain an adjusted large language model assembly;

[0056] Step S4, formulating test content using the adjusted large language model combination;

[0057] Step S5, based on the task scheduling engine 19 and the test content, the evaluation task 7 is assigned to the corresponding test object for testing to obtain result data;

[0058] Step S6, receiving result data, generating a test report 9 and task progress information in combination with the adjusted large language model assembly, and then outputting the result data.

[0059] Specifically, the present invention provides a cloud platform standard compliance evaluation method based on a large language model. For cloud computing standard documents, the large language model technology and full-text retrieval technology are organically combined to realize full-process information management. Based on the large language model, the evaluation task 7 is assigned to the corresponding test object for testing, and the test report and task progress are output to meet the standardized evaluation requirements of cloud computing and realize full-chain traceability.

[0060] In a preferred embodiment, in step S2, the secondary index 14 performs millisecond-level full-text retrieval on the standard document.

[0061] Specifically, elastic search refers to Elasticsearch (an open source distributed search and analysis engine), and the distributed object storage database 13 is specifically minio (an open source high-performance object storage service).

[0062] Build a secondary index based on Elasticsearch14 to achieve millisecond-level full-text search, and provide full-text search capabilities, standard library management, and report management capabilities for upper-level applications.

[0063] The distributed object storage database 13 provides a distributed data storage function to ensure the reliability and accessibility of data. The secondary index 14 is used to improve the efficiency of data query and retrieval. The mongoDB database is used to store and manage related data.

[0064] In a preferred embodiment, the large language model assembly in step S3 includes a Chinese vector model 17 and a Qwen2.5 base large model 15 .

[0065] Specifically, Springcloud microservices provide microservice architecture support to ensure the scalability and flexibility of the system, LoRA efficiently fine-tunes the Qwen2.5 base large model 15 to optimize system performance, and uses the Chinese vector model 17 to process Chinese text and improve the processing capabilities of Chinese data.

[0066] Springcloud is a collection of frameworks that provide microservice architecture support for the entire system. This architecture splits the system into multiple small, independent services, each of which can be independently developed, deployed, and expanded. For example, in the cloud platform standard compliance assessment system, test task scheduling, test case management and other functions can exist as independent microservices. When the system needs to handle more assessment tasks7, the processing capacity can be improved by increasing the number of instances of the corresponding microservices without affecting other functional modules.

[0067] LoRA (Low-Rank Adaptation) efficient parameter fine-tuning is a parameter efficient fine-tuning method. Qwen2.5 base large model 15 is a large language model. After LoRA efficient parameter fine-tuning, it can better handle natural language related tasks in the cloud platform standard compliance evaluation process.

[0068] During the system development process, the base model was fine-tuned for efficient parameters, and experimental comparative analysis was conducted. The evaluation index used in the experiment is ROUGE, where rouge-1 and rouge-2 correspond to the calculation of recall rate after N-gram splitting, and the L in Rouge-L stands for Longest Common Subsequence. The calculation of Rouge-L uses the longest common subsequence (continuous). The experimental data is as follows:

[0069]

[0070] Through experiments, we found that the fine-tuned model has better performance. After improving the data quality, the rouge-L F1 value also exceeded 0.6. Through manual review, most of the predicted instruction results are semantically close to the standard answers.

[0071] For the Chinese vector model 17 (m3e), human feedback reinforcement learning RLHF alignment training was carried out to improve the semantic recognition ability of the Chinese vector model 17 in the cloud computing standard field.

[0072] For example, text data can be collected from various cloud computing standard documents, industry reports, technical specifications, and related cloud computing service operation manuals. These data contain professional terms, concepts, operation procedures, and compliance requirements in the field of cloud computing standards.

[0073] Clean and preprocess the collected data, including removing noise data (such as format errors, irrelevant punctuation marks, etc.), segmenting the text (if it is Chinese data, follow the Chinese segmentation rules), and annotating the data (if necessary, annotate important terms, concepts, etc.).

[0074] The Chinese vector model 17 is used as the basic model. The model has a certain Chinese semantic processing capability in its initial state, but it still needs further optimization for the specific semantics in the field of cloud computing standards.

[0075] The Chinese vector model 17 is used to process data in the field of cloud computing standards to generate candidate answers to specific questions. For example, for the question "What are the data security standards in cloud computing?", the model will generate a series of possible answers based on its understanding of the data.

[0076] Invite experts or professionals in the field of cloud computing to evaluate and provide feedback on the candidate answers generated by the model, pointing out that some important security standards are omitted in the model answers, or that some concepts are not explained accurately.

[0077] Design a reward function based on the feedback. If the model's answer is highly recognized, a higher reward will be given; if the answer contains many errors or inaccuracies, a lower reward or even a penalty will be given.

[0078] Using the reinforcement learning algorithm, the model parameters are updated according to the reward function. By continuously adjusting the parameters, the model gradually learns the answer mode that can obtain higher rewards, that is, the answer mode that is more in line with the semantics of the cloud computing standard field.

[0079] An independent cloud computing standard domain test dataset is used to evaluate the Chinese vector model 17 trained by RLHF. Evaluation indicators include accuracy, recall, F1 value, etc., which are used to measure the performance of the model in semantic recognition.

[0080] If the evaluation results do not meet the expected performance goals, repeat the above feedback reinforcement learning training process to continuously optimize the model parameters until the model's semantic recognition ability in the cloud computing standard field reaches a satisfactory level.

[0081] In a preferred embodiment, referring to Figure 2 , step S6 comprises,

[0082] Step S61, receiving and generating a test report 9 according to result data and a report template 8;

[0083] Step S62, determining task progress information according to the result data;

[0084] Step S63, after the test report 9 is approved by the preset approval strategy, the task progress information and the approved test report 9 are released.

[0085] Specifically, the preset approval policy may be an approval policy based on compliance:

[0086] First, the test results in the test report 9 are compared with the standard documents of the cloud platform. For example, the cloud computing standard stipulates that the security of data storage must reach a certain encryption level.

[0087] If Test Report 9 shows that the data storage encryption level complies with the requirements of the standard document, then the test report is approved on this standard clause; otherwise, if it does not comply, the report will be marked as requiring modification or retesting.

[0088] Set an overall compliance percentage threshold, such as 80%. The system will count the proportion of items that meet the standard in the test report 9 to the total number of test items.

[0089] If the compliance percentage in Test Report 9 reaches or exceeds 80%, Test Report 9 is preliminarily approved; if it is less than 80%, the report needs to be reviewed again, and further analysis of non-compliant items and arrangement of retesting may be required.

[0090] The preset approval policy can also be an approval policy based on data accuracy:

[0091] Check whether the data in the test report 9 is consistent with the original data during the task execution. For example, when testing the network bandwidth performance of the cloud platform, the bandwidth data in the report should be consistent with the data recorded during the actual test.

[0092] If it is found that the reported data deviates from the original data by more than a certain range (such as 5%), the test report 9 will be returned and the data will be required to be re-verified and the report revised.

[0093] Ensure that the test report contains all the required test items and results. For example, for a comprehensive cloud platform standard compliance test, the report should cover test results in multiple aspects such as data processing, network performance, and security compliance. If it is found that the report lacks the results of some key test items, even if the results of other items are qualified, the report cannot be approved and needs to be supplemented and resubmitted.

[0094] In a preferred embodiment, in step S6, the test content, result data and approval strategy are visualized through the low-code engine 16 and the large-screen design engine 18.

[0095] Specifically, the low-code engine 16 uses low-code technology to reduce the amount of code writing and accelerate the development process. The large-screen design engine 18 is used to design and build a large-screen display interface to ensure the intuitiveness and aesthetics of data display.

[0096] The low-code engine 16 is specifically JeecgBoot, which is a tool that can help developers quickly build applications. By providing a visual operation interface and predefined components, it reduces the workload of manual code writing. For cloud platform standard compliance assessment systems, it can accelerate the system development process, especially when frequent adjustments and optimizations of business logic are required.

[0097] The large screen design engine 18 is specifically DataRoom, which is a tool focused on large screen data visualization design. It can display complex data on the large screen in an intuitive and beautiful way, making it convenient for users to analyze data and make decisions. In the standard compliance detection large screen system, it is responsible for effectively visualizing test results, task progress and other data.

[0098] Reference Figure 6 The task scheduling engine 19 performs specific task scheduling operations. The test case visual arrangement arranges the test cases in a visual way to facilitate user operation and management. Code generation based on the arrangement template automatically generates corresponding test codes according to the arrangement template of the test case to improve the test efficiency. Knowledge enhancement retrieval improves the accuracy and efficiency of the retrieval function through knowledge enhancement technology, helping users to better find relevant information.

[0099] Test task scheduling is responsible for scheduling test tasks to ensure that the test process is carried out in an orderly manner. Test case management is used to manage test cases, including the creation, editing, storage and call of test cases. Standard document management is used to manage documents related to standards to ensure the accuracy and completeness of the test basis. Report template management is responsible for managing test report templates to ensure the consistency and standardization of report formats.

[0100] The large screen 10 for standard compliance detection at the top of the entire architecture indicates that the main function of the system is to perform standard compliance detection and display it through the large screen.

[0101] Reference Figure 3 ,Cloud platform standard compliance evaluation system based on large language model, including,

[0102] Storage and retrieval module 1, used for storing standard documents and performing full-text retrieval and management of standard documents;

[0103] The model auxiliary module 2 is connected to the storage and retrieval module 1, obtains the standard document, and adjusts the large language model combination based on the standard document to obtain the adjusted large language model combination;

[0104] Subject module 3, connected to model auxiliary module 2, formulates test content based on the adjusted large language model combination;

[0105] Platform module 4, used to record basic information of the object under test;

[0106] Task module 5, connecting platform module 4 and subject module 3, assigning assessment tasks 7 based on the test content and basic information of the tested object, and then performing the test to obtain result data;

[0107] The result module 6 is connected to the task module 5 and the model auxiliary module 2, receives the result data, generates a test report 9 and task progress information in combination with the adjusted large language model combination, and then outputs it.

[0108] Specifically, the present invention also provides a cloud platform standard compliance assessment system based on a large language model. The subject module 3 mainly solves the problem of what to measure, the platform module 4 mainly solves the problem of who to measure, the task module 5 mainly solves the problem of who to measure, and the result module 6 mainly solves the problem of evaluation. It includes a minio distributed object storage bucket database and a secondary index 14 based on Elasticsearch, which is used to store standard documents and provide efficient full-text retrieval, standard library management and report management capabilities. It is the data storage and retrieval core of the entire system, closely connected with other modules, providing standard document data for the model training module, and providing data support for the result module for report management, etc.

[0109] Task module 5 includes non-interface testing (manual testing) and interface testing, which solves the testing problem of API interfaces not defined in the standard and meets the functional, performance and interface testing requirements of the cloud computing standard.

[0110] An open testing capability has been formed, which can be a scheduling system for people and systems, realize human-machine collaborative operation, and distribute tasks to other testing subsystems, thus releasing the upper limit of testing capabilities.

[0111] In a preferred embodiment, referring to Figure 4 , the storage and retrieval module 1 includes,

[0112] A storage unit 11 stores standard documents via a distributed object storage database 13;

[0113] The retrieval unit 12 is connected to the storage unit 11, and constructs a secondary index 14 based on elastic search to perform full-text retrieval and management on standard documents.

[0114] In a preferred embodiment, the test content includes a set cloud computing subject 31, a data processing subject 32, a metadata subject 33, a network subject 34 and a newly created test subject.

[0115] Specifically, newly created test subjects need to go through the approval process in advance, and the approval process can check whether the newly created test subjects are repeated with existing test subjects. If there are duplications, it may lead to duplication of work and waste of resources.

[0116] For example, the data processing subject may already include testing of data transmission efficiency. If a new test subject is created to test similar aspects of data transmission again, the approval process can detect and avoid such duplication.

[0117] In a preferred embodiment, the platform module 4 enters the basic information of the object under test through low code.

[0118] Specifically, refer to Figure 8 , based on low-code visual orchestration, a new cloud computing virtual private cloud detection component is added, which reads request parameters from a local file, creates a new VPC, and tries to create a new VPC using the read request parameters. If the creation is successful, the ID of the newly created VPC (for example, VPC0001) will be output. If the creation fails, the process will enter the corresponding failure branch.

[0119] For the successfully created VPC (taking VPC0001 as an example), a query operation will be performed. If the query is successful and the queried information is consistent with the creation, both the "Create VPC" and "Query VPC" interfaces are passed. If the query fails, the process will enter the corresponding failure branch.

[0120] If both the Create VPC and Query VPC interfaces fail, the process will obtain another VPC (for example, VPC0002) from the enumeration interface.

[0121] Use the ID of the VPC (VPC0002) obtained from the enumeration interface to perform a query operation. If the query is successful, the "Query VPC" interface is passed. If the query fails, the "Query VPC" interface is not passed. The purpose of enumerating VPCs is to obtain a list of existing VPCs. If the enumeration is successful, the ID of the first VPC in the VPC list (for example, VPC0002) is output. If the enumeration fails, the process will enter the corresponding failure branch.

[0122] Use the VPC (VPC0002) obtained through successful enumeration to perform a query operation. If the query is successful, the "List VPC" operation is successful. If the query fails, the "List VPC" operation is unsuccessful.

[0123] In a preferred embodiment, referring to Figure 5 , Figure 7 , the result module 6 includes,

[0124] A data receiving unit 61, connected to the task module 5, for receiving result data;

[0125] The data combining unit 62 is connected to the data receiving unit 61 and is used to check the result data and obtain the checked result data;

[0126] The report generating unit 63 is connected to the data combining unit 62, and generates a test report 9 for the verified result data according to the preset report template 8 and the adjusted large language model combination;

[0127] The progress confirmation unit 64 is connected to the data compound unit 62 and is used to determine the task progress information according to the verified result data;

[0128] A report approval unit 65, connected to the report generation unit 63, approves the test report 9 according to a preset approval strategy;

[0129] The report publishing unit 66 is connected to the progress confirmation unit 64 and the report approval unit 65 and is used to publish task progress information and the approved test report 9.

[0130] Specifically, the specific test points of subject module 3 are derived from the corresponding industry standards or technical specifications. Based on the big model combined with contextual learning, thinking chain and other prompts, the project generates a large number of test cases and drafts of test items from the standards to assist in writing test cases.

[0131] The test result is a record of the compliance of the corresponding subject of the test object. The test report 9 is generated from the test result and the progress of the test task is determined. During the test report 9 generation process, the subjectivity of the conclusions and suggestions in the report will be given a preliminary draft suggestion by the big model.

[0132] The platform module 4 includes an X platform 41, a Y platform 42, and a Z platform 43. The platform module 4 also supports a low-code approach to designing a platform information entry page, thereby reducing the amount of code written and accelerating the development process.

[0133] The system of the present invention is based on the standard document fine-tuning Chinese vector model 17 and the Lora efficient parameter fine-tuning Qwen2.5 base large model 15, providing semantic vector retrieval capability, intent recognition capability, and retrieval enhancement generation capability for upper-level business applications, and on this basis develops a test case generation assistant and a test report generation assistant; the generated test cases are combined with low-code visual orchestration technology and large model code generation technology to assist in building test components; based on the task scheduling engine 19, detection tasks are assigned to corresponding testers or test subsystems; combined with the JeecgBoot low-code engine 16 and the DataRoom large-screen design engine 18, a visual test case orchestration, visual custom approval process orchestration, and visual large-screen configuration that support drag-and-drop are constructed to form an intelligent, reliable, and interconnected cloud computing standardization construction assessment platform.

[0134] With the help of a large language model, intelligent assistance is achieved in each link, reducing manual intervention, assisting in the evaluation and judgment of pure functions (without API interfaces) based on screenshots, improving efficiency and reducing the error rate. Full life cycle management of each link from standard documents to test logic can not only ensure the continuity of work at each stage, but also realize linkage updates when standard documents change, ensuring that the entire evaluation system keeps pace with the times, always in line with the latest standard requirements, and adapting to the rapid development of the new generation of cloud computing technology.

[0135] In view of the huge and complex architecture of the big data system with cloud computing platform as the core, the system supports distributing test tasks to other subsystems through API authentication to achieve unlimited expansion of test functions. This method can adapt to the test tasks of different components and different levels in the big data architecture, make full use of distributed computing resources, and achieve comprehensive evaluation of complex cloud systems (cloud, data processing, network, security). It has been put into use in the annual assessment of national cloud computing platforms, with the number of man-months invested being 1 / 4 of the previous one and the economic cost being 1 / 6 of the previous one.

[0136] The above description is only a preferred embodiment of the present invention, and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A cloud platform standard compliance evaluation method based on a large language model, characterized in that: include, Step S1, storing the standard document of the cloud platform in a distributed object storage database; Step S2, constructing a secondary index based on elastic search to perform full-text retrieval and management on the standard document; Step S3, adjusting the large language model assembly based on the standard document to obtain the adjusted large language model assembly; Step S4, formulating test content using the adjusted large language model combination; Step S5, based on the task scheduling engine and the test content, assign the evaluation task to the corresponding test object for testing to obtain result data; Step S6, receiving the result data, generating a test report and task progress information in combination with the adjusted large language model assembly, and then outputting the resultant test report and task progress information.

2. The cloud platform standard compliance evaluation method based on a large language model according to claim 1 is characterized in that: The secondary index in step S2 performs millisecond-level full-text retrieval on the standard document.

3. The cloud platform standard compliance evaluation method based on a large language model according to claim 1 is characterized in that: The large language model assembly in step S3 includes a Chinese vector model and a Qwen2.5 base large model.

4. The cloud platform standard compliance evaluation method based on a large language model according to claim 1 is characterized in that: Step S6 comprises, Step S61, receiving and generating the test report according to the result data and the report template; Step S62, determining the task progress information according to the result data; Step S63, after the test report is approved by a preset approval strategy, the task progress information and the approved test report are released.

5. The cloud platform standard compliance evaluation method based on a large language model according to claim 4 is characterized in that: In step S6, the test content, the result data and the approval strategy are visualized through the low-code engine and the large-screen design engine.

6. A cloud platform standard compliance evaluation system based on a large language model, characterized in that: include, A storage and retrieval module, used for storing standard documents and performing full-text retrieval and management of the standard documents; A model auxiliary module is connected to the storage and retrieval module, obtains the standard document, and adjusts the large language model assembly based on the standard document to obtain the adjusted large language model assembly; A subject module, connected to the model auxiliary module, formulates test content according to the adjusted large language model combination; Platform module, used to record basic information of the object under test; A task module, connected to the platform module and the subject module, assigns assessment tasks based on the test content and basic information of the tested object, and then performs the test to obtain result data; The result module is connected to the task module and the model auxiliary module, receives the result data, generates a test report and task progress information in combination with the adjusted large language model assembly, and then outputs it.

7. The cloud platform standard compliance evaluation system based on a large language model according to claim 6 is characterized in that: The storage and retrieval module includes: A storage unit, storing the standard document via a distributed object storage database; The retrieval unit is connected to the storage unit, constructs a secondary index based on elastic search, and performs full-text retrieval and management on the standard document.

8. The cloud platform standard compliance evaluation system based on a large language model according to claim 6 is characterized in that: The test content includes set cloud computing subjects, data processing subjects, metadata subjects, network subjects and newly created test subjects.

9. The cloud platform standard compliance evaluation system based on a large language model according to claim 6 is characterized in that: The platform module enters the basic information of the object under test through low code.

10. The cloud platform standard compliance evaluation system based on a large language model according to claim 6 is characterized in that: The result module includes: A data receiving unit, connected to the task module, for receiving the result data; A data combining unit, connected to the data receiving unit, for checking the result data and obtaining the checked result data; A report generating unit, connected to the data combining unit, generates the test report for the verified result data according to a preset report template and the adjusted large language model combination; A progress confirmation unit, connected to the data combination unit, for determining the task progress information according to the verified result data; A report approval unit, connected to the report generation unit, for approving the test report according to a preset approval strategy; A report publishing unit is connected to the progress confirmation unit and the report approval unit, and is used to publish the task progress information and the approved test report.

Images