Method, device, storage medium and electronic device for determining parameter tuning information

Abstract

The present application discloses a method, device, storage medium and electronic device for determining parameter tuning information. Relating to the field of artificial intelligence, the method comprises: obtaining a first sensitive parameter set of a target application in a current application scenario; judging whether the current application scenario exists in multiple application scenarios based on the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in multiple application scenarios, and obtaining a judgment result; determining a target method from a first method and a second method based on the judgment result, and determining the parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the second method refers to a method of determining parameter tuning information using an algorithm that does not rely on historical tuning data information. The present application solves the technical problem that a single method is used to determine parameter tuning information in the related art, and there is low flexibility in determining parameter tuning information.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method, device, storage medium and electronic device for determining parameter tuning information. Background Art

[0002] There is a need for parameter tuning in scenarios such as operating system kernels, compilers, and databases. Effective parameter settings can optimize service performance and improve overall service quality. However, the number of parameters involved in the above fields is large, which may be as high as hundreds or thousands of dimensions. At the same time, traditional tuning methods often rely on experienced engineers to perform manual tuning based on domain knowledge. High-dimensional parameters pose a great challenge to manual tuning, which may not only lead to low tuning efficiency, but also affect the tuning effect. At present, related technologies usually use a single method to determine parameter tuning information, which results in low flexibility in determining parameter tuning information.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method, device, storage medium and electronic device for determining parameter tuning information, so as to at least solve the technical problem that a single method is used to determine parameter tuning information in the related art, and there is low flexibility in determining parameter tuning information.

[0005] According to one aspect of an embodiment of the present application, a method for determining parameter tuning information is provided, including: obtaining a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; judging whether the current application scenario exists in the multiple application scenarios based on the first sensitive parameter set and a reference sensitive parameter set in the target application scenario in multiple application scenarios, and obtaining a judgment result; determining a target method from a first method and a second method based on the judgment result, and determining the parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining the parameter tuning information using a large language model, the large language model being trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining the parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0006] Furthermore, obtaining a first sensitive parameter set of the target application in the current application scenario includes: obtaining tuning evaluation results corresponding to multiple parameter combinations of the target application in the current application scenario; determining the sensitive parameters of the target application in the current application scenario from the multiple parameter combinations according to the tuning evaluation results corresponding to the multiple parameter combinations through a model interpreter, and obtaining a first sensitive parameter set.

[0007] Furthermore, obtaining tuning evaluation results corresponding to multiple sets of parameter combinations for the target application in the current application scenario includes: determining multiple sets of parameter combinations corresponding to the target application in the current application scenario according to a third method, wherein the third method includes at least one of the following: random selection, Bayesian algorithm, and proxy model; for parameter combinations in the multiple sets of parameter combinations, performing parameter tuning on the target application based on the parameter combination, and performing tuning evaluation on the target application after parameter tuning to obtain tuning evaluation results corresponding to the parameter combination.

[0008] Furthermore, based on the first sensitive parameter set and the reference sensitive parameter set under the target application scenario in multiple application scenarios, it is determined whether the current application scenario exists in the multiple application scenarios, and the judgment result obtained includes: for the target application scenario in the multiple application scenarios, it is determined whether the reference sensitive parameter set of the target application scenario includes the first sensitive parameter set; in the case that there is a reference sensitive parameter set including the first sensitive parameter set, it is determined that the judgment result represents the existence of the current application scenario in the multiple application scenarios; in the case that there is no reference sensitive parameter set including the first sensitive parameter set, it is determined that the judgment result represents the non-existence of the current application scenario in the multiple application scenarios.

[0009] Further, in the case where the target mode is the first mode, determining the parameter tuning information of the target application through the target mode includes: determining a reference sensitive parameter set associated with the current application scenario from reference sensitive parameter sets under multiple application scenarios, and determining the reference sensitive parameter set as a second sensitive parameter set; determining multiple groups of target parameter combinations based on the second sensitive parameter set, and determining tuning evaluation results corresponding to the target application in the multiple groups of target parameter combinations; combining the target parameter combinations in the multiple groups of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain tuning data information, and constructing a first training sample set based on the tuning data information corresponding to the multiple groups of target parameter combinations; training a large language model based on the first training sample set to obtain a target large language model, and determining the parameter tuning information through the target large language model.

[0010] Furthermore, the algorithm that does not rely on historical tuning data information is a Bayesian algorithm. When the target mode is the second mode, determining the parameter tuning information of the target application through the target mode includes: generating an initial parameter combination of the target application in the current application scenario, and determining the tuning evaluation result corresponding to the initial parameter combination; generating an updated parameter combination according to the initial parameter combination and the tuning evaluation result corresponding to the initial parameter combination through the Bayesian algorithm; repeating the step of determining the tuning evaluation result corresponding to the updated parameter combination until the preset iteration condition is met, and determining the parameter tuning information of the target application from the initial parameter combination and the updated parameter combination according to the tuning evaluation result.

[0011] Further, determining the target mode from the first mode and the second mode according to the judgment result includes: if the judgment result indicates that the current application scenario exists among multiple application scenarios, determining the first mode as the target mode; if the judgment result indicates that the current application scenario does not exist among multiple application scenarios, determining the second mode as the target mode.

[0012] According to another aspect of an embodiment of the present application, a method for determining parameter tuning information is also provided, including: obtaining a first sensitive parameter set of a target application uploaded by a client in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; in a cloud server, judging whether the current application scenario exists in multiple application scenarios based on the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in multiple application scenarios, obtaining a judgment result, determining a target method from a first method and a second method based on the judgment result, and determining the parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining the parameter tuning information using a large language model, the large language model being trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining the parameter tuning information using an algorithm that does not depend on historical tuning data information; and feeding back the parameter tuning information of the target application to the client.

[0013] According to another aspect of an embodiment of the present application, a device for determining parameter tuning information is also provided, including: an acquisition unit, used to acquire a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; a judgment unit, used to judge whether the current application scenario exists in multiple application scenarios based on the first sensitive parameter set and a reference sensitive parameter set in the target application scenario in multiple application scenarios, and obtain a judgment result; a determination unit, used to determine a target method from a first method and a second method according to the judgment result, and determine the parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0014] Furthermore, the acquisition unit includes: an acquisition sub-unit, used to obtain tuning evaluation results corresponding to multiple groups of parameter combinations of the target application in the current application scenario; a first determination sub-unit, used to determine the sensitive parameters of the target application in the current application scenario from multiple groups of parameter combinations according to the tuning evaluation results corresponding to the multiple groups of parameter combinations through a model interpreter, and obtain a first sensitive parameter set.

[0015] Furthermore, the acquisition subunit includes: a determination module, used to determine multiple groups of parameter combinations corresponding to the target application in the current application scenario according to a third method, wherein the third method includes at least one of the following: random selection, Bayesian algorithm, and agent model; a processing module, used to perform parameter tuning on the target application based on the parameter combination in the multiple groups of parameter combinations, and perform tuning evaluation on the target application after parameter tuning to obtain a tuning evaluation result corresponding to the parameter combination.

[0016] Furthermore, the judgment unit includes: a judgment subunit, used to judge whether the reference sensitive parameter set of the target application scenario among multiple application scenarios includes a first sensitive parameter set; a second determination subunit, used to determine that the judgment result represents the existence of the current application scenario among multiple application scenarios when there is a reference sensitive parameter set including the first sensitive parameter set; and a third determination subunit, used to determine that the judgment result represents the absence of the current application scenario among multiple application scenarios when there is no reference sensitive parameter set including the first sensitive parameter set.

[0017] Further, in the case where the target mode is the first mode, the determination unit includes: a fourth determination subunit, used to determine a reference sensitive parameter set associated with the current application scenario from reference sensitive parameter sets under multiple application scenarios, and determine the reference sensitive parameter set as the second sensitive parameter set; a fifth determination subunit, used to determine multiple groups of target parameter combinations based on the second sensitive parameter set, and determine the tuning evaluation results corresponding to the target application in the multiple groups of target parameter combinations; a first processing subunit, used to combine the target parameter combinations in the multiple groups of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain tuning data information, and construct a first training sample set according to the tuning data information corresponding to the multiple groups of target parameter combinations; a second processing subunit, used to train a large language model based on the first training sample set to obtain a target large language model, and determine parameter tuning information through the target large language model.

[0018] Furthermore, the algorithm that does not rely on historical tuning data information is a Bayesian algorithm. When the target mode is the second mode, the determination unit includes: a sixth determination subunit, used to generate an initial parameter combination for the target application in the current application scenario, and determine the tuning evaluation result corresponding to the initial parameter combination; a generation subunit, used to generate an updated parameter combination according to the initial parameter combination and the tuning evaluation result corresponding to the initial parameter combination through the Bayesian algorithm; a seventh determination subunit, used to repeatedly execute the step of determining the tuning evaluation result corresponding to the updated parameter combination until a preset iteration condition is met, and determine the parameter tuning information of the target application from the initial parameter combination and the updated parameter combination according to the tuning evaluation result.

[0019] Furthermore, the determination unit includes: an eighth determination subunit, which is used to determine the first method as the target method if the judgment result indicates that the current application scenario exists among multiple application scenarios; and a ninth determination subunit, which is used to determine the second method as the target method if the judgment result indicates that the current application scenario does not exist among multiple application scenarios.

[0020] According to another aspect of an embodiment of the present invention, an electronic device is provided, including: a memory storing an executable program; and a processor for running the program, wherein when the program is running, any one of the above-mentioned methods for determining parameter tuning information is executed.

[0021] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a program, wherein when the program is running, the device where the storage medium is located is controlled to execute any of the above-mentioned methods for determining parameter tuning information.

[0022] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for determining parameter tuning information.

[0023] In an embodiment of the present application, a first sensitive parameter set of a target application in a current application scenario is obtained, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; based on the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in multiple application scenarios, it is determined whether the current application scenario exists in the multiple application scenarios to obtain a judgment result; based on the judgment result, a target method is determined from the first method and the second method, and parameter tuning information of the target application is determined through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0024] In the above process, the large language model is trained based on historical tuning data information in multiple application scenarios, so that the large language model can learn the relevant knowledge in these multiple application scenarios. By judging whether the current application scenario exists in multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, it is realized to accurately judge whether the current application scenario of the target application is an application scenario that has been learned by the large language model according to the feature data corresponding to the current application scenario and the feature data corresponding to multiple application scenarios, that is, to judge whether it is a known application scenario. By determining the target mode from the first mode and the second mode according to the judgment result, it is realized to determine the target mode based on whether the current application scenario is a known application scenario and the dependency relationship between the first mode and the second mode on the data of the application scenario, thereby realizing the dynamic determination of the target mode between the first mode and the second mode according to the actual application situation, so as to further determine the parameter tuning information, thereby effectively improving the flexibility of determining the parameter tuning information. Thus, the purpose of dynamically selecting one mode in the two modes to determine the parameter tuning information is achieved, and the technical effect of effectively improving the flexibility of determining the parameter tuning information is achieved, thereby solving the technical problem that the single mode is used to determine the parameter tuning information in the related technology, and the flexibility of determining the parameter tuning information is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0026] Figure 1 is a schematic diagram of a computer terminal provided according to Embodiment 1 of the present application;

[0027] Figure 2 The process of determining the parameter tuning information according to the first embodiment of the present application is as follows Figure 1 ;

[0028] Figure 3 The process of the optional method for determining parameter tuning information provided in the first embodiment of the present application is as follows Figure 2 ;

[0029] Figure 4 is a flowchart of a method for determining parameter tuning information provided in Embodiment 2 of the present application;

[0030] Figure 5 is a schematic diagram of a device for determining parameter tuning information provided in Embodiment 3 of the present application;

[0031] Figure 6It is a structural block diagram of an electronic device provided according to Embodiment 4 of the present application. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0035] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0036] Parameter tuning: The process of continuously improving the system's score on a benchmark program by adjusting system parameters.

[0037] System parameters: refers to configuration items in the operating system that can be dynamically set and affect application performance, such as kernel configuration parameters and application configuration parameters.

[0038] Benchmark program: An evaluation program used to estimate the performance of a tuning application and give a performance score.

[0039] Parameter tuning algorithm: A program that implements a certain algorithm strategy and continuously gives system parameter combinations to find parameter configurations that improve application performance.

[0040] Proxy model: A mathematical model that constructs a mapping relationship between system parameter combinations and benchmark program scores, and uses this to guide the operation of the parameter tuning algorithm.

[0041] LLM: Large Language Model, large language model.

[0042] Example 1

[0043] According to an embodiment of the present application, a method for determining parameter tuning information is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0044] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for determining parameter tuning information. Figure 1 As shown, the computer terminal (or mobile device) 10 may include a processor set 102 (the processor set 102 may include but is not limited to a processing device such as a microprocessor (Microcontroller Unit, referred to as MCU) or a programmable logic device (Field Programmable Gate Array, referred to as FPGA), and the processor set 102 may include a processor set, Figure 1 102a, 102b, ..., 102n are used to illustrate), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0045] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuits may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device).

[0046] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the method for determining parameter tuning information in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the method for determining the parameter tuning information described above is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0047] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0048] The display may be a touch screen type liquid crystal display, which may enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0049] There are requirements for parameter tuning in scenarios such as operating system kernels, compilers, and databases. Effective parameter settings can optimize service performance and improve overall service quality. However, the number of parameters involved in the above fields is large, which may be as high as hundreds or thousands of dimensions. At the same time, traditional tuning methods often rely on experienced engineers to perform manual tuning based on domain knowledge. High-dimensional parameters pose great challenges to manual tuning, which may not only lead to low tuning efficiency, but also affect the tuning effect.

[0050] There are two main mainstream tuning algorithms at present: algorithms based on Bayesian theory and algorithms based on large language models. The algorithm based on Bayesian theory dynamically searches in the preset parameter space according to the tuning effect of real-time feedback to determine the parameter tuning information. This algorithm framework is relatively general. In recent years, it has not only been used in parameter tuning scenarios related to computer systems, but also widely used in hyperparameter tuning scenarios of deep learning models. When there are many parameters and the dimension of the parameter space is high, the search cycle required by the Bayesian algorithm is often long, resulting in low efficiency. Compared with the Bayesian series of algorithms, the advantage of the LLM-based algorithm is that it can effectively use the historical data of parameter tuning and can be relatively generalized to similar application load fields. At present, the relevant technology usually adopts a single method to determine the parameter tuning information, which leads to the problem of low flexibility in determining the parameter tuning information.

[0051] In the above technical background, the present application provides Figure 2 The method for determining the parameter tuning information shown. Figure 2 The process of determining the parameter tuning information according to the first embodiment of the present application is as follows Figure 1 The method comprises:

[0052] Step S201, obtaining a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree.

[0053] Optionally, electronic devices, application systems, servers and other devices can be used as the execution subject of the present application. In this embodiment, the target processing system is used as the execution subject to execute the method for determining parameter tuning information. Among them, the target application includes but is not limited to database applications, server applications, big data applications, etc. Database applications include but are not limited to relational databases MySQL, Redis (Remote Dictionary Server, remote dictionary service), etc. Server applications include but are not limited to reverse proxy Web (World Wide Web, World Wide Web) servers, and big data applications include but are not limited to applications based on distributed storage and computing frameworks, or open source computing frameworks for large-scale data processing and other frameworks.

[0054] Optionally, the application scenario of the application includes two parts of information: application type and scenario type, where the application type can be MySQL, Redis, etc., and the scenario type can be a read scenario, a write scenario, etc. For example, an optional application scenario can be expressed as "MySQL read scenario". Depending on the actual application requirements, the same application may adopt different scenarios in different time ranges. For example, MySQL is a read scenario during the day, responsible for querying data according to user needs, and a write scenario at night, responsible for updating data in MySQL. Therefore, the same application may correspond to multiple application scenarios. Depending on the actual application requirements, the application type and scenario type are also different, and can be divided at a finer granularity. Therefore, in this embodiment, the content of the application type and scenario type is not specifically limited.

[0055] In an optional embodiment, the target processing system may obtain the first sensitive parameter set of the target application in the current application scenario to achieve effective acquisition of data that can characterize the characteristics of the current application scenario. Optionally, the target processing system may obtain the sensitive parameters input by the user for the current application scenario of the target application, thereby determining the first sensitive parameter set based on the sensitive parameters input by the user. Optionally, the target processing system may also obtain the tuning evaluation results corresponding to multiple groups of parameter combinations of the target application in the current application scenario, and then determine the sensitive parameters of the target application in the current application scenario from the multiple groups of parameter combinations according to the tuning evaluation results corresponding to the multiple groups of parameter combinations through the model interpreter, to obtain the first sensitive parameter set. Optionally, the target processing system may also obtain a group of parameter combinations of the target application in the current application scenario and the tuning evaluation results corresponding to the parameter combination, and input the group of parameter combinations and the tuning evaluation results into the sensitive parameter prediction model to determine the first sensitive parameter set through the sensitive parameter prediction model.

[0056] Step S202: judging whether the current application scenario exists in the multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in the multiple application scenarios, and obtaining a judgment result.

[0057] When an algorithm that does not rely on historical data (such as an algorithm based on Bayesian theory) is used to determine parameter tuning information, its advantage is that it does not require pre-training and fine-tuning, nor does it require the annotation of historical data (i.e., historical tuning data information). Regardless of the scenario, relatively optimal local and locally optimal solutions can be found according to a certain algorithm path. Its disadvantage is that it is impossible to utilize information in historical data, and each time the parameter tuning information is determined, it must start from scratch. It is difficult to find effective parameter tuning information when there are few tuning rounds (at least 2.5 times the number of parameters are required for effective tuning, and more rounds are required in high-dimensional space). In each tuning round, the time required to collect algorithm evaluation indicators (i.e., tuning evaluation results) is much longer than the time the algorithm runs, and the effect of the algorithm based on Bayesian theory in high-dimensional parameter space is similar to random search. Therefore, in high-dimensional space, without adding other algorithms, it is impossible to extract information such as parameter correlation and parameter sensitivity.

[0058] When using a large language model to determine parameter tuning information, its advantage is that it can use historical data to fine-tune the pre-trained initial large language model, thereby effectively reducing the number of tuning rounds in similar scenarios. In addition, during the pre-training and fine-tuning process, in addition to the relationship between parameter combinations and evaluation indicators, it can also extract information such as parameter correlation and parameter sensitivity. Its disadvantage is that the algorithm model is very large, whether it is storage space or the computing and memory resources required for pre-training and fine-tuning are very large, and the running time is long. Whether it is pre-training or fine-tuning, its running time is measured in hours and days. In scenarios without historical data, the results are poor, and it is necessary to rely on other algorithms to generate data for training, and the demand for training data is very large.

[0059] Therefore, if there is already a lot of historical annotated data, and the actual application scenario of the optimization happens to match a scenario in the historical data, then using a large language model to determine the parameter optimization information can effectively reduce the number of runs and obtain good results. Algorithms based on Bayesian theory will perform better in unknown application scenarios (i.e., scenarios without historical data).

[0060] In an optional embodiment, the target processing system can determine whether there is a current application scenario in multiple application scenarios based on the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, that is, determine whether the current application scenario is a known application scenario. The aforementioned multiple application scenarios can be understood as the application scenarios based on which the large language model is trained, that is, the large language model has learned the knowledge in these multiple application scenarios, and these multiple application scenarios are equivalent to known application scenarios.

[0061] Optionally, the target application scenario may be any one of multiple application scenarios.

[0062] Optionally, the reference sensitive parameter set can be pre-stored in a sensitive information library, and the reference sensitive parameter set can be determined by the target processing system through a model interpreter. For example, for a target application scenario among multiple application scenarios, the target processing system can input the historical tuning data information under the target application scenario into the model interpreter, thereby obtaining the influence degree value of the parameters in the historical tuning data information of the target application scenario on the tuning effect applied in the target application scenario, and then the parameters in the historical tuning data information of the target application scenario can be screened based on the influence degree value corresponding to the parameters in the historical tuning data information of the target application scenario to obtain the reference sensitive parameter set corresponding to the target application scenario. For example, the parameters in the historical tuning data information of the target application scenario are sorted in order from high to low according to the influence degree value, and the sorted parameters are obtained, and then the first R parameters in the sorted parameters are determined as sensitive parameters in the reference sensitive parameter set, where R is a positive integer. For another example, the parameters in the historical tuning data information of the target application scenario whose influence degree value is greater than the preset degree value are determined as sensitive parameters in the reference sensitive parameter set.

[0063] Optionally, the reference sensitive parameter set may also be determined by the target processing system through a large language model. For example, in the process of training the large language model, the large language model may learn to obtain reference sensitive parameter sets for different application scenarios based on historical tuning data information for multiple application scenarios. And when the large language model is trained, the large language model outputs reference sensitive parameter sets for different application scenarios.

[0064] Optionally, the historical tuning data information for a certain application scenario includes the parameters, parameter values, and historical tuning evaluation results used when the application performs historical parameter tuning in the application scenario. The real label added to the historical tuning data information for the application scenario may be the application type and scenario type associated with the application scenario, that is, the application scenario is used as the real label of the historical tuning data information. For example, an optional real label may be "MySQL_read_only_full_load", where "MySQL" is equivalent to the application type, and "read_only_full_load" is equivalent to the scenario type of "read-only, full data loading".

[0065] After determining the reference sensitive parameter sets corresponding to the multiple application scenarios, the target processing system may determine whether the current application scenario exists in the multiple application scenarios based on the first sensitive parameter set and the reference sensitive parameter set under the target application scenario in the multiple application scenarios, and obtain a determination result. For example, the target processing system may determine whether the reference sensitive parameter sets corresponding to the multiple application scenarios include a reference sensitive parameter set including the first sensitive parameter set, and thereby determine whether the current application scenario exists in the multiple application scenarios based on the determination result obtained based on the determination.

[0066] For another example, the target processing system can calculate the similarity between the reference sensitive parameter set and the first sensitive parameter set under the target application scenario in multiple application scenarios, thereby judging whether the current application scenario exists in the multiple application scenarios based on the similarity between the reference sensitive parameter set corresponding to the multiple application scenarios and the first sensitive parameter set. For example, if there is a reference sensitive parameter set whose similarity with the first sensitive parameter set is greater than the target threshold, it is determined that the current application scenario exists in the multiple application scenarios, and the application scenario corresponding to the reference sensitive parameter set is the current application scenario. If there is no reference sensitive parameter set whose similarity with the first sensitive parameter set is greater than the target threshold, it is determined that the current application scenario does not exist in the multiple application scenarios. Optionally, the target threshold can be a preset value, and the user can set it according to actual application requirements.

[0067] Step S203, determining a target method from the first method and the second method according to the judgment result, and determining parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0068] Optionally, if the judgment result indicates that the current application scenario exists in multiple application scenarios, the target processing system may determine that the current application scenario is a known application scenario. In this case, the target processing system may determine the first mode as the target mode.

[0069] Optionally, if the judgment result indicates that the current application scenario does not exist in the multiple application scenarios, the target processing system may determine that the current application scenario is an unknown application scenario. In this case, the target processing system may determine the second mode as the target mode.

[0070] In an optional embodiment, the algorithm that does not rely on historical tuning data information may be a Bayesian algorithm.

[0071] In an optional embodiment, the algorithm that does not rely on historical tuning data information may be a random selection algorithm, a proxy model, or other algorithms.

[0072] In an optional embodiment, the parameter tuning information of the target application may be understood as information required for parameter tuning of the target application, and the parameter tuning information may include multiple parameters required for parameter tuning of the target application and parameter values ​​corresponding to the parameters.

[0073] Optionally, when the first method is used as the target method, the target processing system can determine the parameter tuning information of the target application through the large language model. Since the large language model is trained using the labeled historical tuning data information, the large language model can learn the correlation between the application tuning data information and the application scenario of the application, as well as the correlation between the tuning evaluation results and the tuning parameters and parameter values ​​during the training process. Therefore, the target processing system can determine the tuning data information of the target application in the current application scenario, that is, obtain the real-time data of the target application, and then process the tuning data information of the target application in the current application scenario through the large language model to obtain the parameter tuning information of the target application.

[0074] Optionally, when the second method is used as the target method, the target processing system can determine the parameter tuning information of the target application through an algorithm that does not rely on historical tuning data information. For example, the target processing system can determine the tuning parameter information through a Bayesian algorithm. The target processing system can perform parameter tuning on the target application in the current application scenario, and input the parameter combination used in the parameter tuning and the tuning evaluation result obtained after the parameter tuning into the Bayesian algorithm, so as to generate a new parameter combination based on the input data through the Bayesian algorithm, thereby continuously iterating, and when the iteration conditions are met, the parameter tuning information is determined according to the parameter combination used in the iteration process.

[0075] Optionally, after obtaining the parameter tuning information, the target processing system may send the parameter tuning information to the user, so that when receiving the user's application tuning request, the target processing system may perform parameter tuning on the target application based on the parameter tuning information. Optionally, after obtaining the parameter tuning information, the target processing system may also directly perform parameter tuning on the target application based on the parameter tuning information.

[0076] In this solution, by setting the large language model to be trained based on historical tuning data information in multiple application scenarios, the large language model can learn the relevant knowledge in these multiple application scenarios. By judging whether there is a current application scenario in multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, it is realized to accurately judge whether the current application scenario of the target application is an application scenario that has been learned by the large language model according to the feature data corresponding to the current application scenario and the feature data corresponding to multiple application scenarios, that is, to judge whether it is a known application scenario. By determining the target mode from the first mode and the second mode according to the judgment result, it is realized to determine the target mode based on whether the current application scenario is a known application scenario and the dependency relationship between the first mode and the second mode on the data of the application scenario, thereby realizing the dynamic determination of the target mode between the first mode and the second mode according to the actual application situation, so as to further determine the parameter tuning information, thereby effectively improving the flexibility of determining the parameter tuning information. Thus, the purpose of dynamically selecting one mode from the two modes to determine the parameter tuning information is achieved, and the technical effect of effectively improving the flexibility of determining the parameter tuning information is achieved, thereby solving the technical problem that the single mode is used to determine the parameter tuning information in the related technology, and the flexibility of determining the parameter tuning information is low.

[0077] How to accurately obtain data that can characterize the characteristics of the current application scenario is crucial. Therefore, in the method for determining parameter tuning information provided in Example 1 of the present application, obtaining the first sensitive parameter set of the target application in the current application scenario includes: obtaining tuning evaluation results corresponding to multiple groups of parameter combinations of the target application in the current application scenario; determining the sensitive parameters of the target application in the current application scenario from the multiple groups of parameter combinations according to the tuning evaluation results corresponding to the multiple groups of parameter combinations through a model interpreter, and obtaining the first sensitive parameter set.

[0078] In an optional embodiment, the target processing system may first determine multiple groups of parameter combinations corresponding to the target application in the current application scenario, wherein the parameter combinations in the multiple groups of parameter combinations include multiple parameters and parameter values ​​corresponding to the multiple parameters, and the parameters in the parameter combination may be all adjustable parameters of the target application.

[0079] In an optional embodiment, the parameters of the parameter combinations in the multiple groups of parameter combinations are the same, but the parameter values ​​of the parameter combinations are not completely the same.

[0080] For example, for a parameter in the full set of adjustable parameters of the target application in the current application scenario, the target processing system can select any value from the parameter value range corresponding to the parameter as the parameter value of the parameter, thereby generating a set of parameter combinations based on the full set of adjustable parameters and the parameter values ​​corresponding to the parameters. In this way, multiple sets of parameter combinations are obtained by random multiple times.

[0081] Optionally, the parameter value range corresponding to the parameter in the full adjustable parameters can be preset by the user. For example, the target processing system can determine the parameter value range corresponding to the parameter from the full adjustable parameter library. The full adjustable parameter library contains multiple levels and domain parameters of the kernel, kernel driver, compiler, runtime, and application. The number of parameters can be in the range of 2000-100000. The full adjustable parameter library also includes the parameter value range corresponding to the parameter in the full adjustable parameters.

[0082] For another example, the target processing system may use a Bayesian algorithm, or a proxy model, or other algorithms to determine multiple sets of parameter combinations corresponding to the target application in the current application scenario.

[0083] Optionally, the target processing system can determine the tuning evaluation results of the target application in the current application scenario based on the multiple parameter combinations corresponding to the target application. For example, the target application is parameter tuned based on the parameter combinations in the multiple parameter combinations, and the target application after parameter tuning is evaluated to obtain the tuning evaluation results corresponding to the multiple parameter combinations. For another example, for the parameter combinations in the multiple parameter combinations, the target processing system can input the parameter combinations into the target prediction model, and predict the tuning evaluation results corresponding to the parameter combinations through the target prediction model.

[0084] In an optional embodiment, the tuning evaluation result may include multiple evaluation indicators and indicator values ​​corresponding to the evaluation indicators, and the evaluation indicators include but are not limited to response time, throughput, throughput rate, concurrency, throughput bandwidth, etc. The tuning evaluation can be implemented through a benchmark program or other system detection tools.

[0085] Optionally, after obtaining the tuning evaluation results corresponding to multiple sets of parameter combinations of the target application in the current application scenario, for the parameter combination in the multiple sets of parameter combinations, the target processing system can combine the parameter combination and the tuning evaluation results corresponding to the parameter combination to obtain first tuning data information corresponding to the parameter combination.

[0086] Optionally, after obtaining the first tuning data information corresponding to the multiple sets of parameter combinations, the target processing system can input the first tuning data information corresponding to the multiple sets of parameter combinations into the model interpreter, and then use the parameter interpretable algorithm of the model interpreter to determine the influence degree values ​​of the parameters in the multiple sets of parameter combinations on the tuning effect of the target application in the current application scenario according to the first tuning data information. Afterwards, based on the influence degree values ​​corresponding to the parameters in the multiple sets of parameter combinations, the first sensitive parameter set is determined according to the parameters in the parameter combinations. For example, the first sensitive parameter set is determined according to the parameters in the multiple sets of parameter combinations whose influence degree values ​​are greater than the preset degree values.

[0087] Optionally, the parameter interpretability algorithm may be a Lasso (Least Absolute Shrinkage and Selection Operator) algorithm, or an XGboost (eXtreme Gradient Boosting) algorithm, or other algorithms.

[0088] It should be noted that, in the above process, the model interpreter is used to mine the correlation between the parameters in the parameter combination and the tuning evaluation results, thereby effectively improving the accuracy of determining the first sensitive parameter set.

[0089] In order to improve the accuracy of the tuning evaluation results corresponding to the obtained multiple groups of parameter combinations, in the method for determining parameter tuning information provided in Example 1 of the present application, obtaining the tuning evaluation results corresponding to the multiple groups of parameter combinations of the target application in the current application scenario includes: determining the multiple groups of parameter combinations corresponding to the target application in the current application scenario according to a third method, wherein the third method includes at least one of the following: random selection, Bayesian algorithm, and proxy model; for the parameter combinations in the multiple groups of parameter combinations, performing parameter tuning on the target application based on the parameter combination, and performing tuning evaluation on the target application after the parameter tuning to obtain the tuning evaluation results corresponding to the parameter combination.

[0090] Optionally, when the third method is random selection, for a parameter in the full amount of adjustable parameters of the target application in the current application scenario, the target processing system can select any value from the parameter value range corresponding to the parameter as the parameter value of the parameter, thereby generating a set of parameter combinations based on the full amount of adjustable parameters and the parameter values ​​corresponding to the parameters. In this way, multiple sets of parameter combinations are obtained by random selection multiple times.

[0091] Optionally, in the case where the third method is the Bayesian algorithm, the target processing system can perform parameter tuning on the target application in the current application scenario, and input the parameter combination used in the parameter tuning and the tuning evaluation results obtained after the parameter tuning as input data to the Bayesian algorithm, so as to generate new parameter combinations based on the input data through the Bayesian algorithm, thereby iterating preset rounds, and determining the parameter combinations used in the iteration process as the parameter combinations corresponding to the target application in the current application scenario, so as to obtain multiple sets of parameter combinations corresponding to the target application in the current application scenario.

[0092] Optionally, in the case where the third method is a proxy model, the target processing system can first randomly generate multiple groups of random parameter combinations based on the full amount of adjustable parameters of the target application in the current application scenario, and determine the tuning evaluation results corresponding to the multiple groups of random parameter combinations, and then input the multiple groups of random parameter combinations and the tuning evaluation results corresponding to the multiple groups of random parameter combinations as input data into the proxy model, so as to generate new parameter combinations based on the input data through the proxy model, and then determine the tuning evaluation results corresponding to the new parameter combinations, and input the new parameter combinations and their corresponding tuning evaluation results into the proxy model, so as to determine the next new parameter combination through the proxy model, thereby iterating the preset rounds, and determining the parameter combinations generated by the proxy model during the iteration process as the parameter combinations corresponding to the target application in the current application scenario, so as to obtain the multiple groups of parameter combinations corresponding to the target application in the current application scenario.

[0093] Optionally, after determining multiple sets of parameter combinations corresponding to the target application in the current application scenario, the target processing system can perform parameter tuning on the target application based on the parameter combinations in the multiple sets of parameter combinations, and perform tuning evaluation on the target application after parameter tuning to obtain a tuning evaluation result corresponding to the parameter combination.

[0094] It should be noted that by adopting at least one of random selection, Bayesian algorithm, and proxy model to determine the above-mentioned multiple groups of parameter combinations, the flexibility of determining the parameter combination is improved, and by performing parameter tuning based on the parameter combination to determine the tuning evaluation result, the effectiveness of the determined tuning evaluation result is improved, thereby improving the accuracy of determining the first sensitive parameter set, and further improving the accuracy of determining the parameter tuning information.

[0095] In order to accurately determine the target mode, in the method for determining parameter tuning information provided in Example 1 of the present application, determining the target mode from the first mode and the second mode according to the judgment result includes: if the judgment result represents that the current application scenario exists among multiple application scenarios, determining the first mode as the target mode; if the judgment result represents that the current application scenario does not exist among multiple application scenarios, determining the second mode as the target mode.

[0096] Optionally, since the large language model is trained based on historical tuning data information in multiple application scenarios, the large language model has learned relevant knowledge of these multiple application scenarios. Therefore, if the judgment result represents that the current application scenario exists among multiple application scenarios, the target processing system can determine that the current application scenario is a known application scenario. In this case, the large language model can perform effective processing based on the learned knowledge, and the target processing system can determine the first method as the target method to use the large language model to determine parameter tuning information in the known application scenario, thereby reducing tuning rounds to save time and resources.

[0097] Optionally, if the judgment result indicates that the current application scenario does not exist among multiple application scenarios, the target processing system may determine that the current application scenario is an unknown application scenario. In this case, the large language model lacks knowledge of relevant scenarios and the processing effect is relatively poor. The target processing system may determine the second method as the target method to determine the parameter tuning information in the unknown application scenario using an algorithm that does not rely on historical tuning data information (such as a Bayesian algorithm) to avoid affecting the accuracy of determining the parameter tuning information.

[0098] It should be noted that through the above process, it is possible to dynamically select a more suitable algorithm from different algorithms (that is, different methods) based on whether the target application is a known application scenario. Therefore, the use of a large language model in a known application scenario can effectively reduce the number of tuning rounds to save time and resources, while in an unknown application scenario, an algorithm that does not rely on historical tuning data information can be used to ensure that effective parameter tuning information is obtained.

[0099] In order to accurately determine whether the current application scenario is a known application scenario, in the method for determining parameter tuning information provided in Example 1 of the present application, based on the first sensitive parameter set and the reference sensitive parameter set under the target application scenario in multiple application scenarios, it is determined whether the current application scenario exists in multiple application scenarios, and the judgment result obtained includes: for the target application scenario in multiple application scenarios, determining whether the reference sensitive parameter set of the target application scenario includes the first sensitive parameter set; in the case that there is a reference sensitive parameter set including the first sensitive parameter set, determining that the judgment result represents the existence of the current application scenario in multiple application scenarios; in the case that there is no reference sensitive parameter set including the first sensitive parameter set, determining that the judgment result represents the absence of the current application scenario in multiple application scenarios.

[0100] In an optional embodiment, the target application scenario is any one of multiple application scenarios.

[0101] Optionally, for a target application scenario among multiple application scenarios, the target processing system may determine whether the reference sensitive parameter set of the target application scenario includes the first sensitive parameter set. For example, if a certain reference sensitive parameter set includes A parameter, B parameter, and C parameter, and the first sensitive parameter set includes A parameter and B parameter, then it is determined that the reference sensitive parameter set includes the first sensitive parameter set. For another example, if a certain reference sensitive parameter set includes A parameter and C parameter, and the first sensitive parameter set includes A parameter and B parameter, then it is determined that the reference sensitive parameter set does not include the first sensitive parameter set.

[0102] Optionally, when there is a reference sensitive parameter set including the first sensitive parameter set in the first sensitive parameter set corresponding to multiple application scenarios, it is determined that the judgment result represents the existence of the current application scenario among the multiple application scenarios, and the application scenario corresponding to the reference sensitive parameter set including the first sensitive parameter set is determined to be the current application scenario; when there is no reference sensitive parameter set including the first sensitive parameter set in the first sensitive parameter set corresponding to multiple application scenarios, it is determined that the judgment result represents the absence of the current application scenario among the multiple application scenarios.

[0103] It should be noted that by determining whether the current application scenario exists in multiple application scenarios based on the inclusion relationship between the reference sensitive parameter set and the first sensitive parameter set, on the one hand, it is possible to accurately judge whether the current application scenario is a known application scenario, and on the other hand, it effectively improves the judgment efficiency.

[0104] In order to accurately determine the parameter tuning information of the target application, in the method for determining the parameter tuning information provided in Example 1 of the present application, when the target mode is the first mode, determining the parameter tuning information of the target application by the target mode includes: determining a reference sensitive parameter set associated with the current application scenario from reference sensitive parameter sets under multiple application scenarios, and determining the reference sensitive parameter set as a second sensitive parameter set; determining multiple groups of target parameter combinations according to the second sensitive parameter set, and determining the tuning evaluation results corresponding to the target application in the multiple groups of target parameter combinations; combining the target parameter combinations in the multiple groups of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain tuning data information, and constructing a first training sample set according to the tuning data information corresponding to the multiple groups of target parameter combinations; training a large language model based on the first training sample set to obtain a target large language model, and determining the parameter tuning information by the target large language model.

[0105] In an optional embodiment, the reference sensitive parameter sets in multiple application scenarios are learned by the large language model based on historical tuning data information.

[0106] Optionally, the target processing system may determine a reference sensitive parameter set including the first sensitive parameter set among the reference sensitive parameter sets of multiple application scenarios as a reference sensitive parameter set associated with the current application scenario, and determine it as the second sensitive parameter set.

[0107] Optionally, the target processing system may determine multiple groups of target parameter combinations of the target application in the current application scenario according to the second sensitive parameter set, and determine tuning evaluation results corresponding to the multiple groups of target parameter combinations of the target application in the current application scenario.

[0108] In an optional embodiment, in the process of determining multiple groups of target parameter combinations corresponding to the target application according to the second sensitive parameter set, for a sensitive parameter in the second reference sensitive parameter set, any value can be selected from the parameter value range corresponding to the sensitive parameter as the parameter value of the sensitive parameter, thereby generating a group of target parameter combinations according to the sensitive parameters in the second sensitive parameter set and their corresponding parameter values. Thus, multiple groups of target parameter combinations are obtained by random multiple times.

[0109] Optionally, the parameter value range corresponding to the sensitive parameters in the second reference sensitive parameter set may be determined by the target processing system according to the above-mentioned fully adjustable parameter library.

[0110] Optionally, the parameter value range corresponding to the sensitive parameters in the second reference sensitive parameter set can also be determined by the target processing system through a large language model. For example, for a target application scenario in multiple application scenarios, there are multiple historical parameter tunings for the application in the target application scenario, that is, there are historical tuning data information corresponding to multiple historical parameter tunings for the application in the target application scenario. For the same target application scenario, the operating environment of the application in different historical parameter tunings in the target application scenario may be different. Therefore, the large language model can learn the parameters required for parameter tuning in different application scenarios (that is, the reference sensitive parameter set in different application scenarios) and the parameter value range of the parameters in the same application scenario in different operating environments based on the historical tuning data information. Environmental information includes but is not limited to CPU information, memory information, information of the operating system where the application is located, disk space, graphics card information, etc. in the current operating environment of the application in the application scenario. Among them, CPU information can refer to the number of CPU cores, memory information can refer to memory capacity, information of the operating system where the application is located can refer to the system type and version number of the operating system where the application is located, and graphics card information can refer to the graphics card model or performance.

[0111] Therefore, before determining multiple groups of target parameter combinations based on the second sensitive parameter set, the target processing system can also input the environmental information of the current operating environment of the target application and the application type and scenario type associated with the current application scenario into the large language model, so that the large language model can process the environmental information to obtain the parameter value range corresponding to the sensitive parameters in the second sensitive parameter set in the current operating environment of the target application.

[0112] Optionally, after determining multiple groups of target parameter combinations, the target processing system can perform parameter tuning on the target application based on the target parameter combination in the multiple groups of target parameter combinations, and perform tuning evaluation on the target application after parameter tuning to obtain a tuning evaluation result corresponding to the target parameter combination.

[0113] Optionally, the target processing system may combine the target parameter combinations in the multiple sets of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain target tuning data information, and construct a first training sample set based on the target tuning data information corresponding to the multiple sets of target parameter combinations. For example, the target tuning data information corresponding to the multiple sets of target parameter combinations is used as a training sample, and the application type and scenario type associated with the current application scenario are used as true labels to construct a first training sample set. Thus, the first training sample set is input into the large language model to train the large language model based on the first training sample set.

[0114] Optionally, when the target large language model is obtained, the target large language model can automatically output parameter tuning information. For example, when the large language model is trained based on the first training sample set, the target processing system can input a prompt sentence to the large language model to guide the large language model to output parameter tuning information when the target large language model is obtained through training.

[0115] In an optional embodiment, the parameters in the parameter tuning information are the same as the parameters in the target parameter combination, that is, the same as the parameters in the second sensitive parameter set, and the parameter values ​​in the parameter tuning information are determined by the target large language model.

[0116] It should be noted that by determining the first training sample set based on the second sensitive parameter set, it is possible to construct training samples based on parameters that contribute relatively more to application tuning, and by training based on the first training sample set to obtain a target large language model to determine parameter tuning information, it is possible to learn multiple groups of target parameter combinations through the large language model and select parameter information, thereby improving the efficiency and accuracy of determining parameter tuning information.

[0117] In order to obtain a large language model with better performance, in the method for determining parameter tuning information provided in Example 1 of the present application, the large language model learns reference sensitive parameter sets in multiple application scenarios based on historical tuning data information, and the large language model is obtained in the following manner: obtaining historical tuning data information in multiple application scenarios, wherein the historical tuning data information includes parameters, parameter values, and historical tuning evaluation results used when the application performs historical parameter tuning in the application scenarios; using the historical tuning data information in multiple application scenarios as training samples, and the application types and scenario types associated with the application scenarios as true labels to construct a second training sample set; training the initial large language model based on the second training sample set to obtain the large language model.

[0118] In an optional embodiment, the initial large language model can adopt the Transformer algorithm, which is a neural network model based on the self-attention mechanism. The initial large language model can be understood as a pre-trained large language model. The process of obtaining the large language model based on the second training sample set and the process of obtaining the target large language model based on the first training sample set can be understood as a fine-tuning process.

[0119] Optionally, the sensitive parameters in the reference sensitive parameter set corresponding to a certain application scenario refer to parameters whose impact on the tuning effect of the application in the application scenario is greater than the preset target degree. In layman's terms, the sensitive parameters in the reference sensitive parameter set can be understood as the parameters required by the large language model for parameter tuning of the application in the application scenario.

[0120] Optionally, the aforementioned preset degree and the aforementioned preset target degree may be the same or different.

[0121] Optionally, the historical tuning data information in a certain application scenario includes the parameters, parameter values, and historical tuning evaluation results used by the application in the application scenario for historical parameter tuning. The historical tuning evaluation results may be obtained by evaluating the application after the application is historically tuned according to the parameters and parameter values ​​used by the application in historical parameter tuning. The historical tuning evaluation results include multiple evaluation indicators and indicator values ​​corresponding to the evaluation indicators. The evaluation indicators include but are not limited to response time, throughput, throughput rate, concurrency, throughput bandwidth, etc. The tuning evaluation can be implemented through a benchmark program or through other system detection tools.

[0122] After obtaining historical tuning data information under multiple application scenarios, the historical tuning data information under multiple application scenarios can be used as training samples, and the application types and scenario types associated with the application scenarios can be used as true labels to construct a second training sample set, thereby training the initial large language model based on the second training sample set to obtain a large language model.

[0123] In an optional embodiment, it is possible to use Figure 3 The schematic diagram shown implements the training of the initial large language model. Figure 3 The process of the optional method for determining parameter tuning information provided in the first embodiment of the present application is as follows Figure 2 ,like Figure 3As shown, a second training sample set is constructed based on historical tuning data information in multiple application scenarios, and the second training sample set and the full adjustable parameter library are used as inputs of the initial large language model to train the initial large language and obtain a large language model. Among them, the full adjustable parameter library contains multiple levels and domain parameters of the kernel, kernel driver, compiler, runtime, and application, and the number of parameters can be in the range of 2000-100000. By using the full adjustable parameter library as the input of the initial large language model, the large language model can learn the information of all parameters.

[0124] Since the large model is fine-tuned on the basis of the pre-trained LLM, the pre-trained LLM itself has certain knowledge, and with the historical tuning data information used in the fine-tuning process of the large model, the large model can have more preset knowledge in terms of sensitive parameters, value space, parameter correlation, etc. For example, in the process of training the large language model based on the second training sample set, the large language model can learn the parameters required for application parameter tuning in different application scenarios (i.e., the reference sensitive parameter set) and the parameter value range of the parameters applied in the same application scenario in different operating environments based on the historical tuning data information and the real label.

[0125] In an optional embodiment, when the second training sample set is input into the initial large language model, the target processing system may also generate a target instruction sentence and input the target instruction sentence into the initial large language model along with the second training sample set. The target instruction sentence is at least used to guide the initial large language model to output a reference sensitive parameter set in multiple application scenarios learned by the large language model when the large language model is trained. Figure 3 As shown, after the initial large language model is trained based on the second training sample set, a large language model and reference sensitive parameter sets for multiple application scenarios are obtained.

[0126] Afterwards, if Figure 3 As shown, the target processing system can determine the first sensitive parameter set based on some real-time data of the target application in the current application scenario (i.e., the tuning evaluation results corresponding to the multiple sets of parameter combinations of the aforementioned target application in the current application scenario), and thus determine whether the current application scenario is a known application scenario based on the first sensitive parameter set and the reference sensitive parameter set.

[0127] It should be noted that by setting the historical tuning data information including the parameters, parameter values ​​and historical tuning evaluation results used in the historical parameter tuning, the richness of the historical tuning data information is improved. By constructing the second training sample set in the above manner and training the initial large language model based on the second training sample set, the trained large language model can learn the information required for parameter tuning in different application scenarios based on the rich historical data and real labels, thereby improving the working performance of the large language model and further improving the accuracy of determining the parameter tuning information.

[0128] In order to accurately determine the parameter tuning information of the target application, in the method for determining the parameter tuning information provided in Example 1 of the present application, the algorithm that does not rely on historical tuning data information is a Bayesian algorithm. When the target mode is the second mode, determining the parameter tuning information of the target application through the target mode includes: generating an initial parameter combination of the target application in the current application scenario, and determining a tuning evaluation result corresponding to the initial parameter combination; generating an updated parameter combination according to the initial parameter combination and the tuning evaluation result corresponding to the initial parameter combination through the Bayesian algorithm; repeating the step of determining the tuning evaluation result corresponding to the updated parameter combination until a preset iteration condition is met, and determining the parameter tuning information of the target application from the initial parameter combination and the updated parameter combination according to the tuning evaluation result.

[0129] In an optional embodiment, if Figure 3 As shown, when the target processing system determines that the current application scenario is a known application scenario, the large language model is used to determine the parameter tuning information; when the current application scenario is determined to be an unknown application scenario, the Bayesian algorithm is used to determine the parameter tuning information.

[0130] In an optional embodiment, the Bayesian algorithm is a Bayesian algorithm based on Gaussian random process.

[0131] Optionally, the target processing system may generate an initial parameter combination based on the full adjustable parameters of the target application in the current application scenario. For example, according to the parameter value range corresponding to the full adjustable parameters, the parameter values ​​of the parameters in the full adjustable parameters are randomly determined, thereby determining the initial parameter combination based on the full adjustable parameters and the randomly determined parameter values.

[0132] Optionally, after determining the initial parameter combination, the target processing system may perform parameter tuning on the target application in the current application scenario based on the initial parameter combination, and perform tuning evaluation on the target application after parameter tuning, thereby obtaining a tuning evaluation result corresponding to the initial parameter combination.

[0133] Afterwards, the target processing system can input the initial parameter combination and the tuning evaluation results corresponding to the initial parameter combination as input data to the Bayesian algorithm, and then the Bayesian algorithm follows the Gaussian random process, calculates the Gaussian distribution based on the initial parameter combination and the tuning evaluation results corresponding to the initial parameter combination, determines the updated parameter combination in the Gaussian distribution space, and then outputs the updated parameter combination.

[0134] Optionally, the target processing system can determine the tuning evaluation results corresponding to the updated parameter combination, and then input the tuning evaluation results corresponding to the updated parameter combination into the Bayesian algorithm, which generates the next updated parameter combination based on the input, thereby continuously iterating, and when the preset iteration conditions are met, the parameter tuning information of the target application is determined from the initial parameter combination and the updated parameter combination according to the tuning evaluation results.

[0135] In an optional embodiment, the above-mentioned preset iteration condition may refer to the iteration rounds of the Bayesian algorithm reaching the target rounds, and the above-mentioned preset iteration condition may also be that the higher tuning evaluation result determined in the iteration process has not changed or has changed slightly for N consecutive rounds, where N is a positive integer.

[0136] In an optional embodiment, the target processing system may determine a parameter combination with a relatively high tuning evaluation result from the initial parameter combination and the updated parameter combination, thereby determining the group of parameter combinations as parameter tuning information of the target application in the current application scenario.

[0137] For example, the initial parameter combination and the updated parameter combination are sorted in descending order according to the tuning evaluation results to obtain sorted parameter combinations, and then a parameter combination is randomly determined from the first M parameter combinations of the sorted parameter combinations as the parameter tuning information of the target application, where M is a positive integer.

[0138] For another example, the initial parameter combination and the updated parameter combination are sorted in descending order according to the tuning evaluation results to obtain sorted parameter combinations, and then the Pth parameter combination of the sorted parameter combinations is used as the parameter tuning information of the target application, where P is a positive integer.

[0139] It should be noted that, through the above process, the Bayesian algorithm can effectively learn the correlation between parameter combinations and tuning evaluation results, so that accurate parameter tuning information can be determined without relying on historical data, thereby improving the accuracy of determining parameter tuning information.

[0140] In an embodiment of the present application, a first sensitive parameter set of a target application in a current application scenario is obtained, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; based on the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in multiple application scenarios, it is determined whether the current application scenario exists in the multiple application scenarios to obtain a judgment result; based on the judgment result, a target method is determined from the first method and the second method, and parameter tuning information of the target application is determined through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0141] In the above process, the large language model is trained based on historical tuning data information in multiple application scenarios, so that the large language model can learn the relevant knowledge in these multiple application scenarios. By judging whether the current application scenario exists in multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, it is realized to accurately judge whether the current application scenario of the target application is an application scenario that has been learned by the large language model according to the feature data corresponding to the current application scenario and the feature data corresponding to multiple application scenarios, that is, to judge whether it is a known application scenario. By determining the target mode from the first mode and the second mode according to the judgment result, it is realized to determine the target mode based on whether the current application scenario is a known application scenario and the dependency relationship between the first mode and the second mode on the data of the application scenario, thereby realizing the dynamic determination of the target mode between the first mode and the second mode according to the actual application situation, so as to further determine the parameter tuning information, thereby effectively improving the flexibility of determining the parameter tuning information. Thus, the purpose of dynamically selecting one mode in the two modes to determine the parameter tuning information is achieved, and the technical effect of effectively improving the flexibility of determining the parameter tuning information is achieved, thereby solving the technical problem that the single mode is used to determine the parameter tuning information in the related technology, and the flexibility of determining the parameter tuning information is low.

[0142] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0144] Example 2

[0145] According to an embodiment of the present application, a method for determining parameter tuning information is also provided, such as Figure 4 As shown, the method includes:

[0146] Step S401, obtaining a first sensitive parameter set of a target application uploaded by a client in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree.

[0147] Step S402: In the cloud server, based on the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in the multiple application scenarios, it is determined whether the current application scenario exists in the multiple application scenarios, and a judgment result is obtained; based on the judgment result, a target method is determined from the first method and the second method, and parameter tuning information of the target application is determined through the target method, wherein the first method refers to a method of determining parameter tuning information using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios; and the second method refers to a method of determining parameter tuning information using an algorithm that does not depend on historical tuning data information.

[0148] Step S403: Feedback the parameter tuning information of the target application to the client.

[0149] Through the above scheme, the large language model is set to be trained based on the historical tuning data information in multiple application scenarios, so that the large language model can learn the relevant knowledge in these multiple application scenarios. By judging whether the current application scenario exists in multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, it is realized to accurately judge whether the current application scenario of the target application is an application scenario that has been learned by the large language model according to the feature data corresponding to the current application scenario and the feature data corresponding to multiple application scenarios, that is, to judge whether it is a known application scenario. By determining the target mode from the first mode and the second mode according to the judgment result, it is realized to determine the target mode based on whether the current application scenario is a known application scenario and the dependency relationship between the first mode and the second mode on the data of the application scenario, thereby realizing the dynamic determination of the target mode between the first mode and the second mode according to the actual application situation, so as to further determine the parameter tuning information, thereby effectively improving the flexibility of determining the parameter tuning information. Thus, the purpose of dynamically selecting one mode in the two modes to determine the parameter tuning information is achieved, and the technical effect of effectively improving the flexibility of determining the parameter tuning information is achieved, thereby solving the technical problem that the single mode is used to determine the parameter tuning information in the related technology, and the flexibility of determining the parameter tuning information is low.

[0150] In the cloud server, the specific method for determining the parameter tuning information is the same as the method in the first embodiment, and will not be repeated here.

[0151] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0152] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0153] Example 3

[0154] According to an embodiment of the present application, a device for determining parameter tuning information for implementing the above-mentioned method for determining parameter tuning information is also provided. Figure 5 As shown, the device includes: an acquisition unit 501, a judgment unit 502, and a determination unit 503.

[0155] An acquisition unit 501 is used to acquire a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree;

[0156] A determination unit 502 is configured to determine whether the current application scenario exists in the multiple application scenarios according to the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in the multiple application scenarios, and obtain a determination result;

[0157] The determination unit 503 is used to determine the target method from the first method and the second method according to the judgment result, and determine the parameter tuning information of the target application through the target method, wherein the first method refers to a method of determining the parameter tuning information by using a large language model, and the large language model is trained based on historical tuning data information in multiple application scenarios, and the second method refers to a method of determining the parameter tuning information by using an algorithm that does not depend on historical tuning data information.

[0158] In the device for determining parameter tuning information provided in the third embodiment of the present application, the first sensitive parameter set of the target application in the current application scenario is obtained by the acquisition unit 501, wherein the target application is the application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to the parameters whose influence on the tuning effect of the target application in the current application scenario is greater than the preset degree; the judgment unit 502 judges whether the current application scenario exists in multiple application scenarios according to the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in multiple application scenarios, and obtains the judgment result; the determination unit 503 determines the target method from the first method and the second method according to the judgment result, and determines the parameter tuning information of the target application through the target method, wherein the first method refers to the method of determining the parameter tuning information by using a large language model, and the large language model is trained based on the historical tuning data information in multiple application scenarios, and the second method refers to the method of determining the parameter tuning information by using an algorithm that does not depend on the historical tuning data information. In this solution, the large language model is set to be trained based on the historical tuning data information in multiple application scenarios, so that the large language model can learn the relevant knowledge in these multiple application scenarios. By judging whether the current application scenario exists in multiple application scenarios based on the first sensitive parameter set and the reference sensitive parameter set under the target application scenario in multiple application scenarios, it is achieved to accurately judge whether the current application scenario of the target application is an application scenario that has been learned by the large language model based on the feature data corresponding to the current application scenario and the feature data corresponding to multiple application scenarios, that is, to judge whether it is a known application scenario. By determining the target method from the first method and the second method according to the judgment result, it is achieved to determine the target method based on whether the current application scenario is a known application scenario and the dependency relationship between the first method and the second method on the data of the application scenario, thereby achieving the dynamic determination of the target method between the first method and the second method according to the actual application situation, so as to further determine the parameter tuning information, thereby effectively improving the flexibility of determining the parameter tuning information. Thereby, the purpose of dynamically selecting one method from the two methods to determine the parameter tuning information is achieved, and the technical effect of effectively improving the flexibility of determining the parameter tuning information is achieved, thereby solving the technical problem that the single method used in the related technology to determine the parameter tuning information has low flexibility in determining the parameter tuning information.

[0159] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, the acquisition unit 501 includes: an acquisition subunit, used to obtain tuning evaluation results corresponding to multiple groups of parameter combinations of the target application in the current application scenario; a first determination subunit, used to determine the sensitive parameters of the target application in the current application scenario from multiple groups of parameter combinations according to the tuning evaluation results corresponding to the multiple groups of parameter combinations through a model interpreter, and obtain a first sensitive parameter set.

[0160] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, the acquisition subunit includes: a determination module, used to determine multiple groups of parameter combinations corresponding to the target application in the current application scenario according to a third method, wherein the third method includes at least one of the following: random selection, Bayesian algorithm, and agent model; a processing module, used to perform parameter tuning on the target application based on the parameter combination in the multiple groups of parameter combinations, and perform tuning evaluation on the target application after parameter tuning to obtain a tuning evaluation result corresponding to the parameter combination.

[0161] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, the judgment unit 502 includes: a judgment subunit, used to judge whether the reference sensitive parameter set of the target application scenario among multiple application scenarios includes a first sensitive parameter set; a second determination subunit, used to determine that, in the presence of a reference sensitive parameter set including the first sensitive parameter set, the judgment result represents the existence of the current application scenario among multiple application scenarios; and a third determination subunit, used to determine that, in the absence of a reference sensitive parameter set including the first sensitive parameter set, the judgment result represents the absence of the current application scenario among multiple application scenarios.

[0162] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, when the target mode is the first mode, the determination unit 503 includes: a fourth determination subunit, used to determine a reference sensitive parameter set associated with the current application scenario from reference sensitive parameter sets under multiple application scenarios, and determine the reference sensitive parameter set as the second sensitive parameter set; a fifth determination subunit, used to determine multiple groups of target parameter combinations based on the second sensitive parameter set, and determine the tuning evaluation results corresponding to the target application in the multiple groups of target parameter combinations; a first processing subunit, used to combine the target parameter combinations in the multiple groups of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain tuning data information, and construct a first training sample set according to the tuning data information corresponding to the multiple groups of target parameter combinations; a second processing subunit, used to train a large language model based on the first training sample set to obtain a target large language model, and determine the parameter tuning information through the target large language model.

[0163] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, the algorithm that does not rely on historical tuning data information is a Bayesian algorithm. When the target mode is the second mode, the determination unit 503 includes: a sixth determination subunit, used to generate an initial parameter combination for the target application in the current application scenario, and determine a tuning evaluation result corresponding to the initial parameter combination; a generation subunit, used to generate an updated parameter combination according to the initial parameter combination and the tuning evaluation result corresponding to the initial parameter combination through a Bayesian algorithm; a seventh determination subunit, used to repeatedly execute the step of determining the tuning evaluation result corresponding to the updated parameter combination until a preset iteration condition is met, and determine the parameter tuning information of the target application from the initial parameter combination and the updated parameter combination according to the tuning evaluation result.

[0164] Optionally, in the device for determining parameter tuning information provided in Example 3 of the present application, the determination unit 503 includes: an eighth determination subunit, used to determine the first method as the target method if the judgment result indicates that the current application scenario exists among multiple application scenarios; and a ninth determination subunit, used to determine the second method as the target method if the judgment result indicates that the current application scenario does not exist among multiple application scenarios.

[0165] It should be noted that the acquisition unit 501, the judgment unit 502, and the determination unit 503 described above correspond to steps S201 to S203 in Example 1, and the examples and application scenarios implemented by the above units and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0166] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0167] Example 4

[0168] The embodiment of the present application may provide an electronic device, which may be any electronic device in a group of electronic devices. Optionally, in this embodiment, the electronic device may also be replaced by a terminal device such as a mobile terminal.

[0169] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0170] In this embodiment, the electronic device may execute program codes corresponding to the steps in the method for determining parameter tuning information provided in any one of the method embodiments.

[0171] Optionally, Figure 6 is a structural block diagram of an electronic device provided according to Embodiment 4 of the present application. Figure 6 As shown, the electronic device 60 may include: one or more ( Figure 6 Only one is shown in the figure) processor 602, memory 604. The electronic device 60 may also include a storage controller, through which the memory 604 is controlled and managed; the electronic device 60 may also include a peripheral interface, through which the radio frequency module, audio module and display screen are connected.

[0172] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for determining parameter tuning information in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the method for determining the parameter tuning information described above is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0173] The processor may call the information and application programs stored in the memory through the transmission device to execute the program codes corresponding to the steps in the method for determining parameter tuning information provided in any one of the above method embodiments.

[0174] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 6 The structure of the electronic device is not limited. For example, the electronic device 60 may also include Figure 6 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 6 Different configurations are shown.

[0175] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0176] Example 5

[0177] The embodiment of the present application further provides a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the method for determining parameter tuning information provided in the first embodiment.

[0178] Optionally, in this embodiment, the above storage medium may be located in any electronic device in a group of electronic devices in a computer network, or in any mobile terminal in a group of mobile terminals.

[0179] Example 6

[0180] The embodiment of the present application further provides a computer program product. Optionally, in this embodiment, the computer program product may include a computer program, and when the computer program is executed by a processor, the method for determining parameter tuning information provided in the first embodiment is implemented.

[0181] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0182] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0183] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0184] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0185] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0186] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0187] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining parameter tuning information, characterized in that: include: Obtaining a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; According to the first sensitive parameter set and a reference sensitive parameter set under a target application scenario in multiple application scenarios, determining whether the current application scenario exists in the multiple application scenarios, and obtaining a determination result, wherein the target application scenario is any one of the multiple application scenarios; Determine a target mode from the first mode and the second mode according to the judgment result, and determine the parameter tuning information of the target application by the target mode, wherein the first mode refers to a mode of determining the parameter tuning information by using a large language model, and the large language model is trained based on the historical tuning data information in the multiple application scenarios, and the second mode refers to a mode of determining the parameter tuning information by using an algorithm that does not depend on the historical tuning data information; Wherein, determining the target mode from the first mode and the second mode according to the judgment result includes: If the judgment result indicates that the current application scenario exists in the multiple application scenarios, determining the first mode as the target mode; If the judgment result indicates that the current application scenario does not exist in the multiple application scenarios, the second mode is determined as the target mode.

2. The method according to claim 1, characterized in that Obtaining the first sensitive parameter set of the target application in the current application scenario includes: Obtaining tuning evaluation results corresponding to multiple groups of parameter combinations of the target application in the current application scenario; The model interpreter determines the sensitive parameters of the target application in the current application scenario from the multiple parameter combinations according to the tuning evaluation results corresponding to the multiple parameter combinations, so as to obtain the first sensitive parameter set.

3. The method according to claim 2, characterized in that Obtaining the tuning evaluation results corresponding to multiple sets of parameter combinations of the target application in the current application scenario includes: Determine multiple groups of parameter combinations corresponding to the target application in the current application scenario according to a third method, wherein the third method includes at least one of the following: random selection, Bayesian algorithm, and proxy model; For a parameter combination in the plurality of parameter combinations, parameter tuning is performed on the target application based on the parameter combination, and tuning evaluation is performed on the target application after parameter tuning to obtain a tuning evaluation result corresponding to the parameter combination.

4. The method according to claim 1, characterized in that: According to the first sensitive parameter set and a reference sensitive parameter set in a target application scenario among multiple application scenarios, it is determined whether the current application scenario exists in the multiple application scenarios, and the determination result includes: For a target application scenario among the multiple application scenarios, determining whether a reference sensitive parameter set of the target application scenario includes the first sensitive parameter set; In a case where there is a reference sensitive parameter set including the first sensitive parameter set, determining that the judgment result indicates that the current application scenario exists in the multiple application scenarios; In a case where there is no reference sensitive parameter set including the first sensitive parameter set, determining that the judgment result represents that the current application scenario does not exist in the multiple application scenarios.

5. The method according to claim 1, characterized in that: When the target mode is the first mode, determining the parameter tuning information of the target application through the target mode includes: Determine a reference sensitive parameter set associated with the current application scenario from the reference sensitive parameter sets under the multiple application scenarios, and determine the reference sensitive parameter set as a second sensitive parameter set; Determine multiple groups of target parameter combinations according to the second sensitive parameter set, and determine the tuning evaluation results corresponding to the multiple groups of target parameter combinations for the target application; Combining the target parameter combinations in the multiple sets of target parameter combinations with the tuning evaluation results corresponding to the target parameter combinations to obtain tuning data information, and constructing a first training sample set according to the tuning data information corresponding to the multiple sets of target parameter combinations; The large language model is trained based on the first training sample set to obtain a target large language model, and the parameter tuning information is determined through the target large language model.

6. The method according to claim 1, characterized in that The algorithm that does not rely on historical tuning data information is a Bayesian algorithm. When the target mode is the second mode, determining the parameter tuning information of the target application by the target mode includes: Generate an initial parameter combination of the target application in the current application scenario, and determine a tuning evaluation result corresponding to the initial parameter combination; Generate an updated parameter combination according to the initial parameter combination and the tuning evaluation result corresponding to the initial parameter combination by the Bayesian algorithm; Repeat the step of determining the tuning evaluation result corresponding to the updated parameter combination until a preset iteration condition is met, and determine the parameter tuning information of the target application from the initial parameter combination and the updated parameter combination according to the tuning evaluation result.

7. A method for determining parameter tuning information, characterized in that: include: Obtaining a first sensitive parameter set of a target application uploaded by a client in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; In the cloud server, based on the first sensitive parameter set and the reference sensitive parameter set in the target application scenario in the multiple application scenarios, it is judged whether the current application scenario exists in the multiple application scenarios to obtain a judgment result, and the target mode is determined from the first mode and the second mode according to the judgment result, and the parameter tuning information of the target application is determined by the target mode, wherein the first mode refers to a mode of determining the parameter tuning information by using a large language model, and the large language model is trained based on the historical tuning data information in the multiple application scenarios, and the second mode refers to a mode of determining the parameter tuning information by using an algorithm that does not depend on the historical tuning data information, and the target application scenario is any one of the multiple application scenarios, wherein determining the target mode from the first mode and the second mode according to the judgment result includes: if the judgment result indicates that the current application scenario exists in the multiple application scenarios, determining the first mode as the target mode; if the judgment result indicates that the current application scenario does not exist in the multiple application scenarios, determining the second mode as the target mode; Feedback the parameter tuning information of the target application to the client.

8. A device for determining parameter tuning information, characterized in that: include: an acquisition unit, configured to acquire a first sensitive parameter set of a target application in a current application scenario, wherein the target application is an application to be parameter tuned, and the sensitive parameters in the first sensitive parameter set refer to parameters whose influence on the tuning effect of the target application in the current application scenario is greater than a preset degree; a judging unit, configured to judge whether the current application scenario exists in the multiple application scenarios according to the first sensitive parameter set and a reference sensitive parameter set in a target application scenario in the multiple application scenarios, and obtain a judgment result, wherein the target application scenario is any one of the multiple application scenarios; a determination unit, configured to determine a target mode from a first mode and a second mode according to the judgment result, and determine the parameter tuning information of the target application by the target mode, wherein the first mode refers to a mode of determining the parameter tuning information by using a large language model, the large language model is trained based on the historical tuning data information in the multiple application scenarios, and the second mode refers to a mode of determining the parameter tuning information by using an algorithm that does not depend on the historical tuning data information; The determination unit includes: An eighth determining subunit, configured to determine the first mode as the target mode if the judgment result indicates that the current application scenario exists in the multiple application scenarios; A ninth determining subunit is configured to determine the second mode as the target mode if the judgment result indicates that the current application scenario does not exist in the multiple application scenarios.

9. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the method for determining parameter tuning information of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the method for determining parameter tuning information of any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processor, implements the method for determining parameter tuning information according to any one of claims 1 to 7.

Images