Intelligent operation and maintenance method and computing device

Abstract

The embodiment of the invention provides an intelligent operation and maintenance method and computing equipment, which can reduce the learning cost of a user for learning operation of single centralized management software and improve the accuracy of operation and maintenance operation. The method comprises the following steps: receiving operation and maintenance semantic information input by a user, and obtaining an intention recognition result based on the operation and maintenance semantic information; matching target tool information corresponding to the intention recognition result from an operation and maintenance tool set; and calling a target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform, and executing operation and maintenance operation corresponding to the target operation and maintenance tool. Therefore, a user can input the operation and maintenance semantic information corresponding to the operation and maintenance operation to enable the computing device to automatically execute the operation and maintenance operation, and the user does not need to learn a complex operation and maintenance process of the operation and maintenance platform and the corresponding complex operation and maintenance operation.

Description

Technical Field

[0001] This application relates to the technical field of servers, and particularly to an intelligent operation and maintenance method and a computing device. Background Art

[0002] Computing devices usually need to deploy centralized management software (also known as an operation and maintenance platform), and the centralized management software is used to manage operation and maintenance devices. Specifically, the centralized management software can manage the operation and maintenance tasks of operation and maintenance devices by invoking operation and maintenance tools and their corresponding operation and maintenance processes. For example, it can manage the power-off, restart, upgrade, etc. of operation and maintenance devices.

[0003] Currently, the operation and maintenance processes of centralized management software are relatively complex. For example, the centralized management software includes complex operation and maintenance processes such as upgrade processes, log recording processes, alarm processes, or inspection processes. And the operation and maintenance operations corresponding to the operation and maintenance processes often include multiple steps. For example, the upgrade process includes multiple steps such as pre-upgrade preparation, creating an upgrade plan, and version effectiveness. This makes the learning cost of centralized management software relatively high.

[0004] To reduce the learning cost of centralized management software, a guiding method can be adopted currently to guide users to complete complex operation and maintenance operations. For example, a guiding wizard can be set on the guiding interface, and the guiding wizard is used to guide users to manually execute operation and maintenance operations step by step. However, there are differences in the design concepts and operation and maintenance processes of different centralized management software, which results in a relatively high learning cost for users to learn the operations of a single centralized management software, and the operation and maintenance operations are prone to errors. Summary of the Invention

[0005] Embodiments of this application provide an intelligent operation and maintenance method and a computing device, which can reduce the learning cost for users to learn the operations of a single centralized management software and improve the accuracy of operation and maintenance operations.

[0006] In a first aspect, embodiments of this application provide an intelligent operation and maintenance method, which is applied to a computing device. A operation and maintenance platform is deployed on the computing device, and the operation and maintenance platform is used to manage the operation and maintenance operations of operation and maintenance devices. Among them, a pre-processing software is built in the operation and maintenance platform of the computing device or the pre-processing software is independently deployed on the computing device (interacts with the operation and maintenance platform through an application programming interface). The pre-processing software is used to determine operation and maintenance tools, and call the operation and maintenance processes corresponding to the operation and maintenance tools from the operation and maintenance platform, and execute the operation and maintenance operations corresponding to the operation and maintenance processes.

[0007] Specifically, the computing device receives the operation and maintenance semantic information input by the user, and based on the operation and maintenance semantic information, obtains an intent recognition result. The operation and maintenance semantic information is used to represent the semantic information of the operation and maintenance operations triggered by the user. Among them, the intent recognition result includes the operation and maintenance operation type corresponding to the operation and maintenance operation; the target tool information corresponding to the intent recognition result is matched from the operation and maintenance tool set. Among them, the operation and maintenance tool set includes tool information corresponding to multiple operation and maintenance operation types. The tool information is used to trigger the operation and maintenance platform to call the corresponding operation and maintenance tool to execute the corresponding operation and maintenance operation. The computing device calls the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform and executes the operation and maintenance operation corresponding to the target operation and maintenance tool; the operation and maintenance platform is used to store the operation and maintenance tools corresponding to each tool information in the operation and maintenance tool set.

[0008] Thus, the user can input the operation and maintenance semantic information corresponding to the operation and maintenance operation, enabling the computing device to automatically execute the operation and maintenance operation, without the user having to learn the complex operation and maintenance processes and corresponding complex operation and maintenance operations of the operation and maintenance platform. Therefore, it helps to reduce the user's learning cost. And there is no need for the user to manually execute the operation and maintenance operation, so it also helps to improve the accuracy of the operation and maintenance operation.

[0009] In a specific implementation, the computing device generates an intent recognition result according to the operation and maintenance semantic information and the target language model; the target language model is used to process the operation and maintenance semantic information to generate an intent recognition result. The target language model can be used to deeply analyze the operation and maintenance semantic information and accurately and efficiently identify the operation and maintenance intent corresponding to the operation and maintenance operation.

[0010] Among them, the target language model can be a language model obtained by training a general language model based on operation and maintenance corpus; the operation and maintenance corpus indicates a data set related to operation and maintenance operations. By using the corpus associated with the operation and maintenance operations for training, the language model can learn from these corpora and generate context information related to this field, thus significantly improving the efficiency and accuracy of the dialogue system.

[0011] In another specific implementation, the computing device can also perform quantization and compression processing on the target language model so that the storage space occupied by the compressed target language model is less than or equal to a preset storage space threshold. The computing device generates an intent recognition result according to the operation and maintenance semantic information and the compressed target language model. Thus, it is ensured that the language model can run normally on the CPU.

[0012] In another implementation, the computing device can utilize low-rank adaptation fine-tuning technology and operation and maintenance prediction training for a general language model based on a target fine-tuning framework; the target fine-tuning framework is llama-factory. Among them, llama-factory provides a series of tools, interfaces, and methods for fine-tuning the language model, which can simplify the fine-tuning process, reduce the development difficulty, and enable developers to focus more on optimizing the model using operation and maintenance corpus without having to pay too much attention to the underlying complex fine-tuning implementation details. Lora can reduce the computational amount and memory requirements while effectively adapting the model to new tasks and domains by performing low-rank approximation adjustment on some parameters of the model instead of updating all the parameters of the model.

[0013] In a specific example, the computing device converts operation and maintenance semantic information into a query language to obtain target operation and maintenance semantic information; the query language is a language that the computing device can recognize; generates a prompt word corresponding to the target operation and maintenance semantic information, and the prompt word is used to guide the target language model to output content in a specified direction; inputs the prompt word into the target language model to obtain an intention recognition result. After the computing device 10 obtains the target operation and maintenance semantic information, through language conversion and generating a prompt word, the result output by the target language model is made more accurate.

[0014] In another specific implementation, the computing device matches target operation and maintenance operation knowledge from a pre-loaded vector database, and the target operation and maintenance operation knowledge represents operation and maintenance operation knowledge related to the target operation and maintenance semantic information; the vector database pre-stores operation and maintenance operation knowledge based on vector representation; generates a prompt word according to the target operation and maintenance operation knowledge, the target operation and maintenance semantic information, and / or the execution result; the execution result is the result generated after performing the operation corresponding to the operation and maintenance process.

[0015] Optionally, if the intention recognition result meets a preset clarity condition, the computing device matches the target tool information from the operation and maintenance tool set; the preset clarity condition is used to accurately identify the operation and maintenance operation type and the requirements corresponding to the operation and maintenance operation type; if the intention recognition result does not meet the preset clarity condition, re-recognize the operation and maintenance intention corresponding to the operation and maintenance operation to obtain a new intention recognition result.

[0016] In yet another specific implementation, if the new intention recognition result does not meet the preset clarity condition, the computing device records the number of times of re-identification. If the recorded number of times is less than or equal to the preset threshold, re-recognize the operation and maintenance intention corresponding to the operation and maintenance operation until the new intention recognition result meets the preset clarity condition, or until the recorded number of times is greater than the preset threshold, and end the session.

[0017] In a second aspect, an embodiment of the present application provides a computing device, including:

[0018] A memory for storing programs;

[0019] A processor for executing a program stored in a memory, which, when the program stored in the memory is executed, is configured to execute the method according to any one of the first aspect.

[0020] In a third aspect, the present application provides a computer storage medium for storing a computer program, which, when executed, is configured to implement the method provided by any one of the implementation manners in the first aspect of the present application.

[0021] In a fourth aspect, the present application provides a computer program product containing instructions, which, when running on at least one computing device, enables the at least one computing device to implement the method provided by any one of the implementation manners in the first aspect of the present application.

[0022] Any of the intelligent operation and maintenance methods provided above, corresponding computing devices, computer-readable storage media, computer program products, etc. are all configured to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods, which will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 FIG. is a schematic diagram of an application scenario provided by an embodiment of the present application;

[0024] Figure 2 FIG. is a schematic diagram of the hardware structure of a computing device 10 provided by an embodiment of the present application;

[0025] Figure 3A FIG. is a flowchart of an intelligent operation and maintenance method provided by an embodiment of the present application;

[0026] Figure 3B FIG. is a schematic diagram of the display of an operation and maintenance interaction interface provided by an embodiment of the present application;

[0027] Figure 4A FIG. is a flowchart of the implementation of another intelligent operation and maintenance method provided by an embodiment of the present application;

[0028] Figure 4B FIG. is another schematic diagram of an operation and maintenance interaction interface provided by an embodiment of the present application;

[0029] Figure 5 FIG. is another schematic diagram of the specific implementation of an intelligent operation and maintenance method provided by an embodiment of the present application;

[0030] Figure 6 FIG. is a schematic diagram of the structure of an intelligent operation and maintenance device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0032] An embodiment of the present application provides an intelligent operation and maintenance method. By the operation and maintenance semantic information input by the user, the operation and maintenance intention corresponding to the recognition operation and maintenance is executed to obtain an intention recognition result. The target tool information corresponding to the intention recognition result is matched from the operation and maintenance tool set, and the target operation and maintenance tool corresponding to the target tool information is called from the operation and maintenance platform to execute the operation and maintenance operation corresponding to the target operation and maintenance tool. That is, through semantic interaction, the operation and maintenance operation can be executed to manage the operation and maintenance equipment, without the user having to learn the complex operation and maintenance processes and corresponding complex operation and maintenance operations of the operation and maintenance platform. Therefore, it helps to reduce the user's learning cost. And there is no need for the user to manually execute the operation and maintenance operation, so it also helps to improve the accuracy of the operation and maintenance operation.

[0033] First, the application scenario of an intelligent operation and maintenance method provided by the embodiments of the present application will be introduced through the accompanying drawings.

[0034] Exemplarily, as Figure 1 shown, it is a schematic diagram of an application scenario provided by an embodiment of the present application. The computing device 10 is used to manage n operation and maintenance devices (specifically, operation and maintenance device #1 to operation and maintenance device #n). Among them, n is a positive integer.

[0035] In the embodiments of the present application, the operation and maintenance device is a hardware or software entity managed by the computing device, and is used to run the business system and provide services. Among them, the operation and maintenance device is a server, a network device, a storage device, or an Internet of Things device, etc. When the operation and maintenance device is a server, the server can be a blade server, a rack-mounted server, a tower server, a high-density server, or a micro server, etc., and the embodiments of the present application do not specifically limit.

[0036] In the embodiments of the present application, the operation and maintenance platform is deployed on the computing device 10. The operation and maintenance platform is a centralized management software, which is used to uniformly manage the operation and maintenance tasks of multiple operation and maintenance devices, such as restart, power-off, or upgrade, etc. It should be noted that the operation and maintenance tasks correspond to the operation and maintenance processes. For example, if the operation and maintenance task is power-off, the operation and maintenance process is the power-off process; if the operation and maintenance task is upgrade, the operation and maintenance process is the upgrade process.

[0037] It should be noted that in other embodiments, the operation and maintenance platform is also used to monitor the status of the operation and maintenance device in real time, such as monitoring the status of the CPU, memory, disk, or network of the operation and maintenance device, and issuing a warning when an abnormality occurs; managing the configuration information of the operation and maintenance device, etc. The embodiments of the present application do not specifically limit.

[0038] In an embodiment of the present application, the computing device 10 can be a server or other electronic devices such as a laptop computer. When the computing device 10 is a server, the server can be a blade server, a rack server, a tower server, a high-density server or a micro server, etc., and the embodiments of the present application do not specifically limit it. The computing device 10 can be a single server, a server cluster, or a processor in a server, and the embodiments of the present application do not specifically limit it.

[0039] The computing device 10 provided in the embodiments of the present application further includes pre-processing software, which is used to execute the operation of obtaining the operation and maintenance semantic information input by the user, identifying the operation and maintenance intention corresponding to the operation and maintenance semantic information, and obtaining an intention recognition result. For example, the operation and maintenance semantic information input by the user is: power off the server with Internet Protocol (IP) xxx.xxx.xxx.xxx. The pre-processing software identifies the operation and maintenance intention corresponding to the operation and maintenance semantic information, and the obtained intention recognition result is: the power-off process and the operation and maintenance device are the server with the IP address xxx.xxx.xxx.xxx.

[0040] The pre-processing software can match the target tool information corresponding to the intention recognition result from the operation and maintenance tool set. Exemplarily, the intention recognition result obtained by the pre-processing software is: the power-off process and the operation and maintenance device are the server with the IP address xxx.xxx.xxx.xxx. The matched target tool information: power off.

[0041] The pre-processing software calls the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform and executes the operation and maintenance operation corresponding to the target operation and maintenance tool. Exemplarily, the pre-processing software can call a power management tool from the operation and maintenance platform to perform the power-off operation on the server with the IP address xxx.xxx.xxx.xxx.

[0042] Further, the embodiments of the present application further provide a computing device 10 for implementing the intelligent operation and maintenance method provided in other embodiments of the present application.

[0043] Exemplarily, attached Figure 2 FIG. 18 is a schematic hardware structure diagram of a computing device 10 provided in an embodiment of the present application. The computing device 10 may include a processor 110, a memory 120, a universal serial bus (USB) interface 130, a display screen 180, etc.

[0044] Processor 110 may include one or more processing units. For example, processor 110 may include a Central Processing Unit (CPU) and / or a Graphics Processing Unit (GPU), etc. Among them, different processing units may be independent devices or integrated in one or more processors, such as integrated in a System on a Chip (SoC). A memory may also be provided in processor 110 for storing instructions and data. In some embodiments, the memory in processor 110 is a cache memory. This memory can save the instructions or data that processor 110 has just used or recycled.

[0045] Memory 120 can be used to store computer-executable program code, and the executable program code includes instructions. Memory 120 may include a program storage area and a data storage area. Among them, the program storage area can store an operation and maintenance platform, pre-processing software, etc. The data storage area can store the data created during the use of computing device 10 (such as log information, etc.). In addition, memory 120 may include one or more storage units. For example, it may include volatile memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), etc.; it may also include non-volatile memory (NVM), such as read-only memory (ROM), flash memory, etc. Processor 110 executes the intelligent operation and maintenance method of computing device 10 by running the instructions stored in memory 120 and / or the instructions stored in the memory provided in the processor.

[0046] The CPU is the core computing unit of the computing device, used to convert data into digital signals and perform mathematical processing. In the embodiments of the present application, the pre-processing software and the operation and maintenance platform are deployed on the CPU.

[0047] The GPU, also known as the display core, visual processor, display chip, is a microprocessor that specifically performs image and graphics-related computing work on personal computers, workstations, game consoles, and some mobile devices (such as tablets, smartphones, etc.). In the embodiments of the present application, the GPU can process the display of the operation and maintenance interaction page triggered by the user and send the processed operation and maintenance interaction page information to the display for display.

[0048] The display screen 180 is used to display images, videos, etc. The display screen 180 includes a display panel. The display panel can adopt a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active matrix organic light-emitting diode (AMOLED), a flexible light-emitting diode (FLED), a MiniLED, a MicroLED, a Micro-OLED, a quantum dot light-emitting diode (QLED), etc. In some embodiments, the electronic device 100 may include one or N display screens 180, where N is a positive integer greater than 1. In the embodiments of the present application, the display screen 180 is used to display an operation and maintenance interaction page, and the user can input operation and maintenance semantic information on the operation and maintenance interaction page.

[0049] Thus, by understanding the semantic information corresponding to the operation and maintenance operations, the user can make the computing device automatically execute the operation and maintenance operations corresponding to the operation and maintenance operations by inputting the operation and maintenance semantic information, without the user having to learn the complex operation and maintenance processes and corresponding complex operation and maintenance operations of the operation and maintenance platform. Therefore, it helps to reduce the user's learning cost. And there is no need for the user to manually execute the operation and maintenance operations, so it also helps to improve the accuracy of the operation and maintenance operations.

[0050] It should be noted that the above application scenarios and the hardware structure of the computing device are only schematic representations, and in actual use, those skilled in the art can also make adjustments according to needs, and the embodiments of the present application do not specifically limit.

[0051] The following will be described in detail the intelligent operation and maintenance method provided by the embodiments of the present application with reference to the accompanying drawings. To enable those skilled in the art to better understand the intelligent operation and maintenance method provided by the embodiments of the present application, the following will be described by taking the Figure 1 and the Figure 2 shown application scenarios and computing devices as examples.

[0052] The Figure 3A is a flowchart of an intelligent operation and maintenance method provided by the embodiments of the present application, and the method includes the following content:

[0053] S310. The computing device 10 receives an operation and maintenance operation triggered by the user.

[0054] An operation and maintenance operation refers to an operation triggered by a user to convey the user's operation and maintenance intention. An operation and maintenance operation often contains semantic information about the specific operation and maintenance tasks that the user hopes to perform. For example, power off the server with the IP address xxx.xxx.xxx.xxx. The computing device 10 can obtain the user's true operation and maintenance intention through the semantic information contained in the operation and maintenance operation.

[0055] The computing device 10 can receive the operation and maintenance operations triggered by the user in various ways.

[0056] In one example, the display screen of the computing device 10 displays an operation and maintenance interaction interface. The user can trigger an operation and maintenance operation through the operation and maintenance interaction interface.

[0057] Exemplarily, attached Figure 3B is a schematic diagram of the display of an operation and maintenance interaction interface provided by an embodiment of the present application. The operation and maintenance interaction interface is a question-and-answer interface, and operation and maintenance semantic information can be input in the input box of the display interface. For example, the operation and maintenance semantic information input by the user is: collect the alarm information of the server with the IP address xxx.xxx.xxx.xxx. After the display screen obtains the operation and maintenance semantic information input by the user, it sends the operation and maintenance semantic information to the pre-processing software of the CPU for operation and maintenance processing.

[0058] S320. The computing device 10 identifies the operation and maintenance intention corresponding to the operation and maintenance operation to obtain an intention recognition result.

[0059] Identifying the operation and maintenance intention corresponding to the operation and maintenance operation aims to extract the user's operation and maintenance intention from the operation and maintenance operation triggered by the user. Among them, the operation and maintenance intention refers to the operation and maintenance task that the user expects to perform, including the operation and maintenance intention corresponding to the operation and maintenance operation. For example, if the operation and maintenance operation triggered by the user is that the user inputs to collect the alarm information of the server with the IP address xxx.xxx.xxx.xxx, the identified operation and maintenance intention is "collect the alarm information of the server", etc.

[0060] In the embodiment of the present application, the operation and maintenance intention corresponding to the operation and maintenance operation identified by the user is called an intention recognition result.

[0061] After the computing device 10 receives the operation and maintenance operation triggered by the user, it can directly identify the operation and maintenance intention corresponding to the operation and maintenance operation to obtain an intention recognition result.

[0062] Specifically, the computing device 10 can obtain the operation and maintenance semantic information input by the user. Then, preprocess and extract features from the operation and maintenance semantic information. Convert the operation and maintenance semantic information into features that can be processed by a machine learning model or a deep learning model. Among them, the preprocessing can include, but is not limited to, word segmentation, removing stop words, and / or punctuation marks, etc.

[0063] Next, the computing device 10 inputs the extracted features into the corresponding learning model, and through the processing of the learning model, an intention recognition result is obtained. Among them, the learning model can be a traditional machine learning model such as a support vector machine model or a Bayesian classifier, or a deep learning model such as a recurrent neural network or a long short-term machine. The embodiments of the present application do not specifically limit this.

[0064] S330. The computing device 10 matches the target tool information corresponding to the intention recognition result from the operation and maintenance tool set.

[0065] The operation and maintenance tool set is a tool set pre-stored in the computing device 10 for calling external tools to perform specific tasks or operations. In the embodiments of the present application, the operation and maintenance tool set includes tool information corresponding to various operation and maintenance operation types. For example, the operation and maintenance operation types include the first tool information corresponding to the log management operation, the second tool information corresponding to the alarm operation, the third tool information corresponding to the inspection operation, and the fourth tool information corresponding to the upgrade operation, etc.

[0066] Among them, the tool information is used to trigger the operation and maintenance platform to call the corresponding operation and maintenance tool to perform the corresponding operation and maintenance operation. In the embodiments of the present application, the tool information corresponds to the operation and maintenance tool. Exemplarily, if the tool information is the first tool information, the operation and maintenance tool is the log tool; if it is the second tool information, the operation and maintenance tool is the alarm tool; if it is the third tool information, the operation and maintenance tool is the inspection tool; if it is the fourth tool information, the operation and maintenance tool is the upgrade tool. Among them, the operation and maintenance tool is used to perform the corresponding operation and maintenance operation. For example, the inspection tool is used to regularly check the system status to determine that the system is running normally, and the alarm tool is used to trigger an alarm to notify relevant personnel when the system has an abnormality or specific conditions, etc.

[0067] In the embodiments of the present application, after the computing device 10 obtains the intention recognition result, it can match the target tool information corresponding to the intention recognition result from the operation and maintenance tool set. In one example, the computing device 10 can first parse and identify specific keywords in the intention recognition result, and use the specific keywords to match the target tool information corresponding to the intention recognition result from the operation and maintenance tool set. For example, the computing device 10 parses and identifies the specific keyword in the intention recognition result as log management. Then the computing device 10 can match the first tool information corresponding to the log management operation from the operation and maintenance tool set.

[0068] It should be noted that the computing device 10 can also match the target tool information from the operation and maintenance tool set through other means, and the embodiments of the present application do not specifically limit this.

[0069] S340. The computing device 10 calls the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform and performs the operation and maintenance operation corresponding to the target operation and maintenance tool.

[0070] After the computing device 10 obtains the target tool information, the pre-processing software can call the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform and execute the operation and maintenance operation corresponding to the target operation and maintenance tool.

[0071] Among them, the operation and maintenance platform pre-stores the operation and maintenance tools corresponding to each tool information in the operation and maintenance tool set, and / or the operation and maintenance processes corresponding to the operation and maintenance tools. By calling the operation and maintenance tools, the operation and maintenance platform can execute the corresponding operation and maintenance operations based on the operation and maintenance processes corresponding to the operation and maintenance tools.

[0072] Thus, when the user understands the operation and maintenance operation, by triggering the operation and maintenance operation, the computing device can automatically execute the operation and maintenance operation corresponding to the operation and maintenance operation, without the user having to learn the complex operation and maintenance processes and corresponding complex operation and maintenance operations of the operation and maintenance platform. Therefore, it helps to reduce the user's learning cost. And there is no need for the user to manually execute the operation and maintenance operation, so it also helps to improve the accuracy of the operation and maintenance operation.

[0073] Embodiment 2

[0074] Furthermore, the computing device 10 can use the target language model to identify the operation and maintenance intention corresponding to the operation and maintenance operation and obtain the intention recognition result. The target language model is a lightweight large language model and can operate efficiently on Figure 2 the CPU described above. The following will be described in detail with reference to the accompanying Figure 4A drawings.

[0075] Accompanying Figure 4A is a flowchart of the implementation of another intelligent operation and maintenance method provided by the embodiment of the present application.

[0076] Among them, step S410 in the accompanying Figure 4A drawings is the same as S310 shown in the accompanying Figure 3A drawings, Figure 3A S320 in the accompanying Figure 3A drawings can be refined into S420 to S460, S330 in the accompanying Figure 4A drawings is the same as S470 in the accompanying Figure 3A drawings, and S340 in the accompanying Figure 4A drawings is refined into S480 and S490 in the accompanying

[0077] As Figure 4A shown, the intelligent operation and maintenance method specifically includes the following content:

[0078] S410. The computing device 10 receives the operation and maintenance operation triggered by the user.

[0079] Among them, the operation and maintenance semantic information corresponding to the operation and maintenance operations received by the computing device 10 is natural language. Natural language conforms to people's daily communication habits, does not require professional programming syntax or instruction formats, and meets the needs of non-professionals to express operation and maintenance tasks. Therefore, it helps to improve the universality of this intelligent operation and maintenance algorithm.

[0080] For example, the operation and maintenance semantic information is: power off the server with the IP of 192.168.1.1.

[0081] S420. The computing device 10 converts the operation and maintenance semantic information into a query language to obtain the target operation and maintenance semantic information.

[0082] Due to the rich expression of natural language, different people may express the same operation and maintenance task differently. For example, for the operation and maintenance task of powering off a server, the corresponding natural language can be "power off that server", or "perform a power-off operation on this server", etc. And natural language may also be ambiguous, that is, there may be ambiguity or inaccurate expressions. This makes it difficult for the computing device 10 to directly process natural language, and the processing results are inaccurate. Therefore, the computing device 10 first converts the operation and maintenance semantic information from natural language into a query language.

[0083] Among them, the query language is a structured query language that can be understood by the computing device and has a fixed format. In the embodiments of the present application, the query language includes an operation and maintenance operation type and an operation and maintenance object identifier. For example, the query language is {"action":, "server_ip":}. Among them, the "action" field indicates the operation and maintenance operation type, and the "server_ip" field indicates the operation and maintenance object identifier. For the operation and maintenance semantic information in natural language "power off the server with the IP of xxx.xxx.xxx.xxx", the generated query format is {"action": "shutdown", "server_ip": "xxx.xxx.xxx.xx"}. Among them, shutdown means power off.

[0084] In the embodiments of the present application, the operation and maintenance semantic information in the query language is called the target operation and maintenance semantic information.

[0085] After the computing device 10 obtains the target operation and maintenance semantic information, it can clearly guide the operation and maintenance object and the operation and maintenance operation type. Therefore, it helps to improve the response accuracy of the computing device 10 and makes the automated operation and maintenance more efficient and accurate.

[0086] (Optionally) S430. The computing device 10 loads the vector database and matches the target operation and maintenance operation knowledge corresponding to the target operation and maintenance semantic information from the vector database.

[0087] The vector database, also known as the knowledge base, is used to store historical knowledge documents, past cases, and historical operation manuals related to the operation and maintenance field. In the embodiments of this application, the vector database includes pre-stored operation and maintenance knowledge represented by vectors.

[0088] The computing device 10 can load the vector database and match the target operation and maintenance operation knowledge corresponding to the target operation and maintenance semantic information from it. For example, if the target operation and maintenance semantic information is {"action": "shutdown", "server_ip": "xxx.xxx.xxx.xx"}, the computing device 10 can retrieve knowledge related to the server power-off operation from the vector database. For example, the target operation and maintenance operation knowledge includes: the standard process of the power-off operation; matters that need to be checked before the server is powered off (such as saving data, closing running programs, etc.); the command to execute the power-off operation, etc.

[0089] S440. The computing device 10 generates a prompt word according to the target operation and maintenance operation knowledge, the target operation and maintenance semantic information, and / or the execution result.

[0090] The prompt word is used to guide the target language model to output content in a specified direction. In the embodiments of this application, the prompt word refers to words suitable for processing by the language model. Before the large language model processes, the computing device 10 first generates a prompt word, which can better interact with the language model and improve the recognition accuracy.

[0091] In one example, the computing device 10 can generate a prompt word according to the target operation and maintenance semantic information.

[0092] However, this way of only considering the target operation and maintenance semantic information has a single factor, and the generated prompt word lacks practicality and pertinence (that is, the quality of the generated prompt word is low). Therefore, the computing device 10 can comprehensively consider the target operation and maintenance operation knowledge and the target operation and maintenance semantic information, and clearly indicate the operation and maintenance operation requirements of the server and the target operation and maintenance operation knowledge in the prompt word. For example, the generated prompt word can incorporate the specific steps of the power-off operation in sequence. This makes

[0093] In one example, the computing device 10 can generate a prompt word according to the target operation and maintenance operation knowledge, or can generate a prompt word according to the target operation and maintenance semantic information. The prompt word is no longer a broad and general expression, but is customized closely around the operation and maintenance operation to be performed at present, and can effectively guide the language model or other execution modules to carry out corresponding operation and maintenance activities more accurately. That is, comprehensively considering the target operation and maintenance operation knowledge and the target operation and maintenance semantic information can effectively improve the pertinence and practicality of the prompt word in assisting to complete specific operation and maintenance tasks.

[0094] Furthermore, the execution result refers to the result information generated by performing an operation and maintenance operation, including but not limited to feedback information, intermediate results, etc. For example, the execution result is that there is missing or inaccurate information in the preliminary operation and maintenance operation generated previously.

[0095] In another example, the computing device 10 can comprehensively consider the target operation and maintenance operation knowledge, the target operation and maintenance semantic information, and the execution result to generate a prompt word. This way of generating a prompt word by adding feedback on the execution result can further improve the quality and accuracy of the generated prompt word.

[0096] It should be noted that in the embodiments of the present application, the computing device 10 can also separately consider the target operation and maintenance operation knowledge to generate prompt words, etc., and the embodiments of the present application do not specifically limit this.

[0097] Exemplarily, if the target operation and maintenance semantic information is {"action": "shutdown", "server_ip": "xxx.xxx.xxx.xx"}, and the obtained target operation and maintenance operation knowledge is: the standard process of the power-off operation; matters that need to be checked before the server is powered off (such as saving data, closing running programs, etc.); the command to execute the power-off operation, etc., then the generated prompt word can be the following content:

[0098] {"Please perform a shutdown operation on the server with IP address xxx.xxx.xxx.xx. Follow the standard shutdown procedure and ensure all data is saved before proceeding}。

[0099] S450. The computing device 10 inputs the prompt word into the target language model to obtain an intention recognition result.

[0100] The target language model is used to perform recognition processing on the prompt word to obtain the operation and maintenance intention corresponding to the prompt word, that is, to obtain the intention recognition result.

[0101] In the embodiments of the present application, the target language model can be a large language model pre-trained based on operation and maintenance corpus. Among them, the operation and maintenance corpus refers to a data set related to operation and maintenance operations, including but not limited to: operation and maintenance operation instructions, fault reports and handling records, system logs, operation and maintenance documents, etc.

[0102] In one example, the target language model can be a large language model obtained by training a general language model with operation and maintenance corpus. Among them, the general language model is open source and a pre-trained language model.

[0103] Exemplarily, the computing device can train the general language model based on the target fine-tuning framework, using the target fine-tuning technology and operation and maintenance corpus. Among them, the target fine-tuning framework can provide a series of tools and interfaces for fine-tuning the large language model. For example, the model framework can be llama-factory, which provides a series of tools, interfaces and methods for fine-tuning the language model, can simplify the fine-tuning process, reduce the development difficulty, and enable developers to focus more on optimizing the model using the operation and maintenance corpus without having to pay too much attention to the underlying complex fine-tuning implementation details.

[0104] The embodiment of the present application can integrate model training on the CPU of the computing device 10. Since training a large language model requires a large amount of computing power and memory requirements, which will affect the operation of the CPU, the target fine-tuning technology can be Low-Rank Adaptation (Lora). Lora can reduce the amount of computation and memory requirements by performing low-rank approximation adjustment on some of the model's parameters instead of updating the parameters of the entire model, and at the same time can effectively make the model adapt to new tasks and domains.

[0105] Specifically, through llama-factory, the operation of passing the general language model and the operation and maintenance corpus into the Lora fine-tuning model can significantly improve the semantic parsing ability of the general language model in the operation and maintenance field.

[0106] Furthermore, the general language model provided by the embodiment of the present application can be a lightweight large language model. Compared with non-lightweight language models, lightweight language models occupy less actual memory and take less time. Exemplarily, for the first language model and the second language model, where the scale of the first language model is smaller than that of the second language model, Tables 1 to 3 below show the actual memory and time consumption corresponding to the output annotation (Token) length of the first language model and the actual memory and time consumption corresponding to the output Token length of the second language model.

[0107] Table 1

[0108] Output Token Length CPU Configuration Actual CPU Used Memory Configuration [[ID= ​ 32 2C 200％ 2G ​ 5～7 64 2C 200％ 2G ​ 10～14 128 2C 200％ 2G ​ 28～29 256 2C 200％ 2G ​ 54～56 512 2C 200％ 2G ​ 97～116 1024 2C 200％ 2G ​ 339～240

[0109] Table 2

[0110] ​ ​ ​ ​ ​ ​ 32 2C 200％ 2G ​ 11-13 64 2C 200％ 2G ​ 20-24 128 2C 200％ 2G ​ 39-46 256 2C 200％ 2G ​ 88-90 512 2C 200％ 2G ​ 178-181 1024 2C 200％ 2G ​ 280-393

[0111] By comparing Table 1 and Table 2, it can be determined that the first language model occupies less actual memory and takes less time. It can be understood that selecting a lightweight language model as the general language model can reduce the amount of memory occupied by the actual CPU and improve the response speed.

[0112] Furthermore, the target language model can also be a language model after quantization compression by a quantization compression algorithm, and the actual storage space occupied (also known as the memory amount) is less than or equal to a preset storage space threshold. The preset storage space threshold refers to the maximum memory amount that satisfies the normal operation of the CPU. For example, the preset storage space threshold is 8GB. Thus, it is ensured that the language model can operate normally on the CPU.

[0113] The quantization compression algorithm can be a low-precision quantization algorithm, that is, converting model parameters and calculations from high precision, such as 32-bit floating-point numbers, to low precision, such as 16-bit floating-point numbers. It can also be other quantization compression algorithms, which are not specifically limited in the embodiments of the present application.

[0114] Exemplarily, Table 3 is a parameter comparison table of the first language model and the second language model provided by the embodiments of the present application after compression. Exemplarily, the precision of fp16 is compressed to INT4. Among them, FP16: A floating-point number is represented by 16-bit binary numbers, including 1 sign bit, 5 exponent bits, and 10 mantissa bits. INT4: An integer is represented by 4-bit binary numbers, and the value range is generally -7 to 7 (including the sign) or 0 to 15 (unsigned).

[0115] Table 3

[0116] ​ ​ ​ ​ ​ ​ ​ ​ ​ 1.2G ​ 685M 200％ 21s ​ ​ 469M ​ 669M 200％ 5-7s ​ ​ 1.2G ​ 685M 400％ 5s ​ ​ 469M ​ 669M 400％ 2.69s

[0117] Comparing Table 3, it can be found that for the language model after quantization compression, the model is smaller, the memory occupancy is reduced, and it can operate normally on the CPU. In addition, when processed using this language model, the processing time will also be significantly reduced.

[0118] The target language model of the embodiments of the present application can also be subjected to other processes, which are not specifically limited in the embodiments of the present application.

[0119] It should be noted that the general language model provided by the embodiments of the present application can be the qwen2.5 model or other models, which are not specifically limited in the embodiments of the present application.

[0120] S460. The computing device 10 determines whether the intent recognition result meets a preset clarity condition. If it meets, S470 is executed; if not, optionally, S440 is executed.

[0121] Considering that some intention recognition results are clear, while some are not. Exemplarily, the matching degree between the intention recognition result and the operation and maintenance semantic information is relatively low. For example, the operation and maintenance semantic information is "power off the server with IP xxx.xxx.xxx.xxx", while the intention recognition result is power off, but the key information of the server with IP address "xxx.xxx.xxx.xxx" is not highlighted. The unclear intention recognition result will affect the subsequent operation and maintenance effect and accuracy. In the embodiments of the present application, clear conditions are first preset to prepare for recognizing the operation and maintenance operation type and the requirements corresponding to the operation and maintenance operation type. For example, the preset clear conditions include but are not limited to: the matching degree between the intention recognition result and the operation and maintenance operation is higher than the preset matching degree threshold, the expression is unambiguous and / or the expression is clear, etc.

[0122] In the embodiments of the present application, if the intention recognition result meets the preset conditions, perform operation S470. If the intention recognition result does not meet the preset clear conditions, S440 can be re-executed to reprocess and obtain the intention recognition result.

[0123] To avoid unlimited execution of intention recognition, the computing device 10 can also set a threshold for re-recognition (i.e., the preset threshold). If the number of re-recognition times is greater than the preset threshold, the session operation is ended. Specifically, the computing device 10 includes a counter for the number of recognition times. If the new intention recognition result does not meet the preset clear conditions, the computing device 10 records the number of re-recognition times. If the recorded number of recognition times is less than or equal to the preset threshold, the operation and maintenance intention corresponding to the operation and maintenance operation is re-recognized. If the recorded number of recognition times is greater than the preset threshold, the session is ended.

[0124] S470: The computing device 10 identifies the target tool information corresponding to the intention recognition result from the operation and maintenance tool set.

[0125] S480: The computing device 10 calls the operation and maintenance interface to obtain the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform.

[0126] Exemplarily, in the embodiments of the present application, the target operation and maintenance tool information can be displayed on the operation and maintenance interaction interface.

[0127] Exemplarily, attached ​ FIG. is another schematic diagram of the operation and maintenance interaction interface provided by the embodiments of the present application. The operation and maintenance interaction interface is a question-and-answer interface, which includes the operation and maintenance semantic information input by the user and displays the target operation and maintenance tool information. ​ Taking the operation and maintenance operation input by the user as: collect the alarm information of the IP xxx.xxx.xxx.xxx, and the displayed target operation and maintenance tool information is "I need to call the collect-log tool" as an example for illustrative description.

[0128] Accordingly, the user can determine whether the invoked operation and maintenance tool meets the requirements based on the operation and maintenance interaction interface, thereby improving the execution accuracy of operation and maintenance operations.

[0129] S490. The computing device 10 executes the operation and maintenance operation corresponding to the target operation and maintenance tool.

[0130] Optionally, S4100. The computing device 10 generates an execution result and executes S440.

[0131] For the convenience of understanding the intelligent operation and maintenance method shown in the appendix ​ The embodiments of the present application provide a specific implementation of the intelligent operation and maintenance method. Appendix ​ FIG. is a schematic diagram of another specific implementation of the intelligent operation and maintenance method provided by the embodiments of the present application. This method is applied to the preprocessing software, and the preprocessing software interacts with the operation and maintenance platform through the operation and maintenance interface.

[0132] The method includes the following steps:

[0133] S501. The preprocessing software receives an operation and maintenance task.

[0134] That is, it receives the operation and maintenance operation triggered by the user and obtains the operation and maintenance semantic information corresponding to the operation and maintenance operation.

[0135] S502. Perform language conversion.

[0136] Specifically, the preprocessing software converts the operation and maintenance semantic information into a query language to obtain the target operation and maintenance semantic information.

[0137] S503. Load the knowledge base.

[0138] Specifically, the preprocessing software reads the vector database and matches the target operation and maintenance operation knowledge corresponding to the target operation and maintenance semantic information from the vector database.

[0139] S504. Generate a prompt

[0140] S505. Invoke the target language model for processing

[0141] The preprocessing software inputs the prompt into the target language model for processing and outputs an intention recognition result.

[0142] S506. Analyze the intention recognition result

[0143] S507. If the intention is not clear and the number of attempts to recognize is less than or equal to the preset threshold, trigger S504 until the intention recognition is clear or the number of attempts to recognize is greater than the preset threshold, and then execute S508.

[0144] If the intention is clear, the preprocessing software automatically calls tool information to invoke external functional functions from the operation and maintenance tool set. For example, the invoked external functional functions include logging, alarming, inspection, management, deployment, and upgrade. Then, the preprocessing software calls the operation and maintenance interface to interact with the operation and maintenance platform, invokes the corresponding operation and maintenance tool from the operation and maintenance interface, executes the operation and maintenance operation, and generates an execution result.

[0145] Exemplarily, the following code is an example of a specific implementation for invoking the operation and maintenance interface to obtain an operation and maintenance tool. Among them, the operation and maintenance tool is specifically illustrated by the power-off tool (i.e., PowerOffTool). This code specifically includes:

[0146] class PowerOffTool

[0147] name = "power_off"

[0148] description = "Use this tool to turn off the device power"

[0149] def _run(self, tool_input: str) -> str:

[0150] # Try to extract the IP address from the input

[0151] match = re.search(r'\regex expression\b', tool_input)

[0152] ip = match.group(0) if match else None

[0153] print("Preparing to turn off the device power...")

[0154] confirmation = input("Do you want to continue to turn off the device power? (yes / no):")

[0155] if confirmation.lower() == 'yes':

[0156] if not ip:

[0157] ip = input("Please enter the IP address of the device:")

[0158] sn = input("Please enter the SN number of the device:")

[0159] password = input("Please enter the password:")

[0160] print(f"Start simulating the call to the interface: IP: {ip}, device SN: {sn}, password: {password}")

[0161] return "The device power has been turned off."

[0162] else:

[0163] return "The operation has been cancelled."

[0164] In the above code, class represents a class, and classPowerOffTool represents defining a class named PowerOffTool. name is the name attribute of the class, with the value "power_off", which is used to describe the name of the operation and maintenance tool. description: The description attribute of the class, with the value "Use this tool to turn off the device power", which is used to describe the purpose of the operation and maintenance tool.

[0165] def_run is a function named def_run. This function includes an input parameter (tool_input), which is a string (i.e., str). This function attempts to extract the IP address from the input parameter. ->str indicates that the function returns a string representing the operation result. For example, the return result is "The operation has been cancelled". self is an instance of the class, representing the object that calls the method.

[0166] match is a storage field used to store the result of the regular expression match. Its value is re.search(r'\regular expression\b', tool_input). Among them, re.search is the search function, and r'\regular expression\b' is a regular expression pattern. re.search(r'\regular expression\b', tool_input) means the function to search for a match of the regular expression from the input parameter.

[0167] ip = match.group(0) if match else None indicates that if the match result is not found (i.e., match else), it means the match fails and returns None (None means the return fails). If the match is successful, it returns match.group(0), where match.group(0) represents returning the entire matched string.

[0168] print refers to the output field, with the value ("Preparing to turn off the device power..."), which outputs "Preparing to turn off the device power".

[0169] confirmation is a variable whose value is either "yes" or "no", indicating whether to continue powering off the device. If the value is "yes", it means to continue powering off the device; otherwise, stop powering off the device. The input() function indicates where the input is obtained from the user.

[0170] confirmation.lower() == 'yes' means converting the string in confirmation to lowercase "yes".

[0171] if not ip means if ip is empty or None

[0172] ip = input("Please enter the IP address of the device:") means prompting the user to enter the IP address of the device and copying the entered value to the variable ip.

[0173] sn = input("Please enter the SN number of the device:") means prompting the user to enter the serial number (Serial Number, SN) of the device and copying the entered value to the variable sn.

[0174] password = input("Please enter the password:") means prompting the user to enter the password and copying the entered password to the variable password.

[0175] print(f"Start simulating calling the interface: IP: {ip}, device SN: {sn}, password: {password}") means outputting the simulated call to the interface, where the IP for calling the interface is the variable ip, the device SN is the variable sn, and the password is the variable password.

[0176] return "The device power has been turned off" means returning that the device power has been turned off.

[0177] The pre - processing software feeds back the execution result to regenerate the prompt word according to the execution result, target operation and maintenance semantic information, and target operation and maintenance operation knowledge, that is, execute S504.

[0178] The pre - processing software is an open - source framework that can directly process the target language model, specifically including the following processing steps:

[0179] S501 - 1. Model quantization and compression.

[0180] S501 - 2. Model fine - tuning.

[0181] It should be noted that the execution order of S501 - 1 and S501 - 2 can be adjusted by itself.

[0182] In summary, the embodiments of the present application can efficiently obtain the operation and maintenance intention of users by using a language model. Further, more accurate prompt words that better conform to the operation and maintenance intention of users can be generated through target operation and maintenance operation knowledge, target operation and maintenance semantic information, and execution results. Therefore, by processing the prompt words using the language model, the user intention can be accurately recognized. Moreover, the language model adopted in the embodiments of the present application is a general language model trained with operation and maintenance corpus and is a compressed model, which can reduce the consumption of computing resources on the premise of ensuring the recognition accuracy.

[0183] Furthermore, the embodiments of the present application also provide an intelligent operation and maintenance device, which is applied to a computing device.

[0184] Appendix ​ FIG. is a schematic structural diagram of an intelligent operation and maintenance device provided by an embodiment of the present application. The device 600 includes:

[0185] A receiving unit 601, configured to receive the operation and maintenance semantic information input by the user, where the operation and maintenance semantic information is used to represent the semantic information of the operation and maintenance operation triggered by the user;

[0186] An identification unit 602, configured to obtain an intention recognition result based on the operation and maintenance semantic information; the intention recognition result includes the operation and maintenance operation type corresponding to the operation and maintenance operation;

[0187] A matching unit 603, configured to match the target tool information corresponding to the intention recognition result from the operation and maintenance tool set, where the operation and maintenance tool set includes tool information corresponding to multiple operation and maintenance operation types; the tool information is used to trigger the operation and maintenance platform to call the corresponding operation and maintenance tool to execute the corresponding operation and maintenance operation;

[0188] A calling unit 604, configured to call the target operation and maintenance tool corresponding to the target tool information from the operation and maintenance platform of the computing device to execute the operation and maintenance operation corresponding to the target operation and maintenance tool; the operation and maintenance platform is used to store the operation and maintenance tools corresponding to each tool information in the operation and maintenance tool set.

[0189] Optionally, the identification unit 602 is specifically configured to: generate an intention recognition result according to the operation and maintenance semantic information and the target language model; the target language model is used to process the operation and maintenance semantic information to generate the intention recognition result.

[0190] Wherein, the target language model is a language model obtained after training a general language model based on operation and maintenance corpus; the operation and maintenance corpus indicates a data set related to operation and maintenance operations.

[0191] Optionally, the device 600 further includes a training unit, configured to perform quantization compression processing on the target language model, so that the storage space occupied by the compressed target language model is less than or equal to a preset storage space threshold;

[0192] The recognition unit 602 is further configured to generate an intent recognition result based on the operation and maintenance semantic information and the compressed target language model. The computing device may train a universal language model using low-rank adaptive fine-tuning techniques and operation and maintenance predictions based on a target fine-tuning framework, which is a llama-factory.

[0193] Optionally, the recognition unit 602 is specifically used to: convert the operation and maintenance semantic information into a query language to obtain target operation and maintenance semantic information; the query language is a language that can be recognized by the computing device; generate prompt words corresponding to the target operation and maintenance semantic information, and the prompt words are used to guide the target language model to output content in a specified direction; input the prompt words into the target language model to obtain intent recognition results.

[0194] Optionally, the identification unit 602 is also used to: match target operation and maintenance knowledge from a pre-loaded vector database, the target operation and maintenance knowledge representing the operation and maintenance knowledge related to the target operation and maintenance semantic information; the vector database pre-stores the operation and maintenance knowledge based on vector representation; generate prompt words based on the target operation and maintenance knowledge, the target operation and maintenance semantic information and / or the execution result; the execution result is the result generated after executing the operation and maintenance operation corresponding to the operation and maintenance process.

[0195] Optionally, the matching unit 603 is specifically used to: if the intention recognition result meets the preset clarity conditions, match the target tool information from the operation and maintenance tool set; the preset clarity conditions are used to accurately identify the operation and maintenance operation type and the corresponding requirements of the operation and maintenance operation type; if the intention recognition result does not meet the preset clarity conditions, re-identify the operation and maintenance intention corresponding to the operation and maintenance operation to obtain a new intention recognition result.

[0196] Optionally, the apparatus 600 further includes a circulation unit, specifically configured to:

[0197] If the new intent recognition result does not meet the preset clarity conditions, the number of re-identifications is recorded. If the recorded number of recognitions is less than or equal to the preset threshold value, the operation and maintenance intent corresponding to the operation and maintenance operation is re-identified until the new intent recognition result meets the preset clarity conditions, or until the recorded number of recognitions is greater than the preset threshold value, the session is ended.

[0198] Thus, through the above device, users understand the operation and maintenance operations and can trigger the operation and maintenance operations to automatically execute the corresponding operation and maintenance operations on the computing device. This eliminates the need for users to learn the complex operation and maintenance processes of the operation and maintenance platform and the corresponding complex operation and maintenance operations, thus helping to reduce user learning costs. Furthermore, since users do not need to manually perform the operation and maintenance operations, this also helps to improve the accuracy of the operation and maintenance operations.

[0199] The embodiments of the present application also provide a computer program product containing instructions. The computer program product can be software or a program product containing instructions that can run on a computing device or be stored in any available medium. When the computer program product runs on a computing device, it causes the computing device to execute the above-mentioned intelligent operation and maintenance method. The embodiments of the present application also provide a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state drive), etc. The computer-readable storage medium includes instructions that instruct the computing device to execute the above-mentioned intelligent operation and maintenance method.

[0200] The descriptions of the processes or structures corresponding to the above respective drawings have their own focuses. For parts not detailed in a certain process or structure, reference can be made to the relevant descriptions of other processes or structures.

[0201] As described above, the above are only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent operation and maintenance method, characterized in that, Applied to a computing device, the method includes: Receiving operation and maintenance semantic information input by a user, where the operation and maintenance semantic information is used to represent the semantic information of an operation and maintenance operation triggered by the user; Based on the operation and maintenance semantic information, obtaining an intention recognition result; the intention recognition result includes an operation and maintenance operation type corresponding to the operation and maintenance operation; Matching target tool information corresponding to the intention recognition result from an operation and maintenance tool set, where the operation and maintenance tool set includes tool information corresponding to multiple operation and maintenance operation types; the tool information is used to trigger the operation and maintenance platform to call a corresponding operation and maintenance tool to execute a corresponding operation and maintenance operation; Invoking, from the operation and maintenance platform of the computing device, the target operation and maintenance tool corresponding to the target tool information to execute the operation and maintenance operation corresponding to the target operation and maintenance tool; the operation and maintenance platform is used to store the operation and maintenance tools corresponding to each tool information in the operation and maintenance tool set.

2. The method according to claim 1, characterized in that, The obtaining the intention recognition result based on the operation and maintenance semantic information includes: Generating the intention recognition result according to the operation and maintenance semantic information and a target language model; the target language model is used to process the operation and maintenance semantic information to generate the intention recognition result.

3. The method according to claim 2, wherein The target language model is a language model obtained by training a general language model based on operation and maintenance corpus; the operation and maintenance corpus indicates a data set related to operation and maintenance operations.

4. The method according to claim 2 or 3, characterized in that, The method further includes: Performing quantization compression processing on the target language model so that the storage space occupied by the compressed target language model is less than or equal to a preset storage space threshold; The generating the intention recognition result according to the operation and maintenance semantic information and the target language model includes: generating the intention recognition result according to the operation and maintenance semantic information and the compressed target language model.

5. The method according to claim 2, characterized in that, The training the general language model based on operation and maintenance corpus includes: training the general language model based on a target fine-tuning framework, using low-rank adaptation fine-tuning technology and the operation and maintenance corpus; the target fine-tuning framework is llama-factory.

6. The method according to claim 2, wherein The generating the intention recognition result according to the operation and maintenance semantic information and the target language model includes: Converting the operation and maintenance semantic information into a query language to obtain target operation and maintenance semantic information; the query language is a language recognizable by the computing device; Generating a prompt corresponding to the target operation and maintenance semantic information, where the prompt is used to guide the target language model to output content in a specified direction; Inputting the prompt into the target language model to obtain the intention recognition result.

7. The method according to claim 6, wherein The generating the prompt corresponding to the target operation and maintenance semantic information includes: Matching target operation and maintenance operation knowledge from a pre-loaded vector database, where the target operation and maintenance operation knowledge represents operation and maintenance operation knowledge related to the target operation and maintenance semantic information; the vector database pre-stores operation and maintenance operation knowledge based on vector representation; Generating the prompt according to the target operation and maintenance operation knowledge, the target operation and maintenance semantic information, and / or an execution result; the execution result is a result generated after executing the operation and maintenance operation corresponding to the operation and maintenance process.

8. The method according to claim 1, wherein The matching the target operation and maintenance tool corresponding to the intention recognition result from the operation and maintenance tool set includes: If the intention recognition result meets the preset clarity condition, match the target tool information from the operation and maintenance tool set; the preset clarity condition is used to accurately identify the operation and maintenance operation type and the requirements corresponding to the operation and maintenance operation type. The method further includes: if the intention recognition result does not meet the preset clarity condition, re-recognize the operation and maintenance intention corresponding to the operation and maintenance operation to obtain a new intention recognition result.

9. The method according to claim 8, wherein After obtaining the new intention recognition result, the method further includes: If the new intention recognition result does not meet the preset clarity condition, record the number of times of re-recognition. If the recorded number of times of re-recognition is less than or equal to the preset threshold, re-recognize the operation and maintenance intention corresponding to the operation and maintenance operation until the new intention recognition result meets the preset clarity condition, or until the recorded number of times of re-recognition is greater than the preset threshold, and end the session.

10. A computing device, characterized in that, It includes a processor and a memory, and the processor is coupled to the memory; The memory is used to store programs; The processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1-9.

Images