Multi-path recall retrieval method and device for computing power resource operation data of intelligent computing center

Abstract

The invention provides a multi-path recall retrieval method and device for computing power resource operation data of an intelligent computing center, and the method comprises the steps: S1, obtaining a user question which is used for querying the computing power resource operation data of the intelligent computing center, the user question comprises at least two of the following retrieval information: vector retrieval information, document retrieval information and scalar retrieval information; s2, analyzing the user question to obtain each piece of retrieval information in the user question; s3, a database is retrieved through the at least two kinds of retrieval information to obtain computing power resource operation data retrieval results matched with the at least two kinds of retrieval information, and operation data of a plurality of computing power resources are stored in the database; and S4, fitting the computing power resource operation data retrieval results matched with the at least two kinds of retrieval information to obtain a final computing power resource operation data retrieval result. According to the method, the retrieval result of the computing power resource operation data is more accurate and more comprehensive.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure technologies, and particularly relates to a method and device for retrieving multiple paths of operation data of computing power resources in an intelligent computing center. Background Art

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "intelligent computing centers" have emerged as the times require.

[0003] An "intelligent computing center" refers to a facility that uses large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, to mainly provide the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios like artificial intelligence deep learning model development, model training, and model inference). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0004] The "intelligent computing center" includes, but is not limited to, the "intelligent computing center".

[0005] An "intelligent computing center", that is, an artificial intelligence computing center, is a type of computing power infrastructure based on artificial intelligence theory, adopting an artificial intelligence computing architecture, and providing computing power services, data services, and algorithm services required for artificial intelligence applications.

[0006] "Computing power" is the core of "intelligent computing centers" and "intelligent computing centers". It is the ability of computer devices or computing / data centers to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of target results through processing information data, and a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, which mainly provides services to society through computing power infrastructure.

[0007] Since the emergence of intelligent computing centers, querying and retrieving the operation data of computing power resources in intelligent computing centers has been an urgent problem to be solved. An intelligent computing center contains a large number of computing power resources, such as computing resources like CPUs (Central Processing Units), GPUs (Graphics Processing Units), network resources like switches and routers, storage resources like storage arrays and distributed storage, security resources like firewalls and intrusion detection systems, as well as support and guarantee resources like wind, fire, water, and electricity. The operation data volume of these computing power resources is huge, and the storage method is relatively complex. For example, the operation data of computing power resources may be stored in vector mode, document mode, and / or scalar mode simultaneously. Therefore, it is difficult to query and retrieve the operation data of computing power resources in intelligent computing centers, and thus it is difficult to finely manage and control computing power resources. Summary of the Invention

[0008] The present invention provides a method and device for retrieving operation data of computing power resources in an intelligent computing center in multiple ways, which are used to solve the problem of difficult query and retrieval of operation data of computing power resources in the intelligent computing center.

[0009] To solve the above technical problems, the present invention is implemented as follows:

[0010] In a first aspect, the present invention provides a method for retrieving operation data of computing power resources in an intelligent computing center in multiple ways, including:

[0011] Step S1: Obtain a user question for querying the operation data of computing power resources in the intelligent computing center, where the user question includes at least two of the following retrieval information: vector retrieval information, document retrieval information, and scalar retrieval information;

[0012] Step S2: Analyze the user question to obtain at least two types of retrieval information in the user question;

[0013] Step S3: Retrieve the database through the at least two types of retrieval information respectively to obtain a retrieval result of operation data of computing power resources that matches the at least two types of retrieval information, where the database stores the operation data of multiple computing power resources;

[0014] Step S4: Fit the retrieval result of operation data of computing power resources that matches the at least two types of retrieval information to obtain a final retrieval result of operation data of computing power resources.

[0015] Optionally, step S3 includes:

[0016] Step S31: If the retrieval information includes vector retrieval information, convert the vector retrieval information into a target vector, calculate the similarity between the target vector and the vector data of each computing power resource in the database, and obtain a first retrieval result according to the calculation result. The first retrieval result includes the computing power resources corresponding to the top N vector data with the highest similarity;

[0017] Step S32: If the retrieval information includes document retrieval information, decompose the document retrieval information into multiple morphemes, calculate the correlation score between each morpheme and the document data of each computing power resource in the database, and determine the total correlation score of the document data according to the correlation scores of all morphemes and the document data. Obtain a second retrieval result according to the total correlation score. The second retrieval result includes the computing power resources corresponding to the top M document data with the highest total correlation score;

[0018] Step S33: If the retrieval information includes scalar retrieval information, query the database to obtain a third retrieval result. The third retrieval result includes the computing power resources that match the scalar retrieval information.

[0019] Optionally, step S4 includes:

[0020] Step S41: If the computing power resource operation data retrieval result only includes the first retrieval result and the second retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the second retrieval result, and perform a weighted sum of the similarity and the total relevance score of the computing power resources with the same identifier to obtain the top X computing power resources with the highest weighted sum score as the final computing power resource operation data retrieval result;

[0021] Step S42: If the computing power resource operation data retrieval result includes the first retrieval result, the second retrieval result, and the third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result, the second retrieval result, and the third retrieval result, and perform a weighted sum of the similarity and the total relevance score of the computing power resources with the same identifier to obtain the top Y computing power resources with the highest weighted sum score as the final computing power resource operation data retrieval result;

[0022] Step S43: If the computing power resource operation data retrieval result includes the first retrieval result and the third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the third retrieval result, and obtain the top Z computing power resources with the highest similarity among the computing power resources with the same identifier as the final computing power resource operation data retrieval result;

[0023] Step S44: If the computing power resource operation data retrieval result includes the second retrieval result and the third retrieval result, obtain the computing power resources with the same identifier in the second retrieval result and the third retrieval result, and obtain the top S computing power resources with the highest total relevance score among the computing power resources with the same identifier as the final computing power resource operation data retrieval result.

[0024] Optionally, before step S41 or step S42, it further includes:

[0025] Step S40: Analyze the user's question to obtain the weights of the similarity and the total relevance score; or, obtain the pre-set weights of the similarity and the total relevance score.

[0026] Optionally, it further includes:

[0027] Step S5: Monitor the computing power resource operation data of the intelligent computing center in real time;

[0028] Step S6: Record the monitored computing power resource operation data into the database.

[0029] Optionally, the computing power resources include at least one of the following: computing resources, network resources, storage resources, security resources, and support and guarantee resources;

[0030] The operation data of the computing power resources includes at least one of the following: basic performance index data, power consumption and power management data, video memory related data, heat dissipation and fan control data, hardware status and error monitoring data, process and task management data, performance optimization index data, interface and data transmission data, driver and system information, advanced function monitoring data, tool and scenario application data.

[0031] In a second aspect, the present invention provides a multi-way recall retrieval device for the operation data of the computing power resources of an intelligent computing center, including:

[0032] An acquisition module, configured to acquire a user question, where the user question is used to query the operation data of the computing power resources of the intelligent computing center, and the user question includes at least two of the following retrieval information: vector retrieval information, document retrieval information, and scalar retrieval information;

[0033] An analysis module, configured to analyze the user question to obtain at least two retrieval information in the user question;

[0034] A retrieval module, configured to respectively retrieve a database through the at least two retrieval information to obtain a retrieval result of the operation data of the computing power resources that matches the at least two retrieval information, where multiple operation data of the computing power resources are stored in the database;

[0035] A fitting module, configured to fit the retrieval result of the operation data of the computing power resources that matches the at least two retrieval information to obtain a final retrieval result of the operation data of the computing power resources.

[0036] In a third aspect, the present invention provides an electronic device, including: a processor, a memory, and a program stored on the memory and executable on the processor, where when the program is executed by the processor, the steps of the multi-way recall retrieval method for the operation data of the computing power resources of the intelligent computing center as described in the first aspect above are implemented.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, where when the computer program is executed by a processor, the steps of the multi-way recall retrieval method for the operation data of the computing power resources of the intelligent computing center as described in the first aspect above are implemented.

[0038] In a fifth aspect, the present invention provides a computer program product, including computer instructions, where when the computer instructions are executed by a processor, the steps of the multi-way recall retrieval method for the operation data of the computing power resources of the intelligent computing center as described in the first aspect above are implemented.

[0039] In the present invention, by analyzing the input user question, at least two of the vector retrieval information, document retrieval information, and scalar retrieval information in the user question are obtained. The database is queried through multiple paths to obtain the retrieval results of the computing power resource operation data matching each retrieval information, and the retrieval results of the computing power resource operation data matching each retrieval information are fitted to obtain the final retrieval results of the computing power resource operation data, solving the problem of difficult query and retrieval of the computing power resource operation data in the intelligent computing center. Due to the adoption of the multi-path recall retrieval method, the retrieval results of the computing power resource operation data are more accurate and comprehensive, realizing the precise management of the computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0041] Figure 1 is a schematic flow chart of the multi-path recall retrieval method for the computing power resource operation data of the intelligent computing center of the present invention;

[0042] Figure 2 is a schematic diagram of the multi-path recall retrieval method for the computing power resource operation data of the intelligent computing center according to an embodiment of the present invention;

[0043] Figure 3 is a schematic structural diagram of the multi-path recall retrieval device for the computing power resource operation data of the intelligent computing center of the present invention;

[0044] Figure 4 is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The technical solutions in the present invention will be clearly and completely described below with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] The terms related to the present invention will be described below.

[0047] The "computing power" as described in the present invention refers to: the ability of a computer device or a computing / data center to process information, which is the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, and is the computing ability to achieve the output of a target result through the processing of information data. It is a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0048] The "computational power" (Computational Power, CP) as described in the present invention refers to: the ability of a data center server to process data and achieve result output, which is a comprehensive indicator for measuring the computing ability of a data center and includes general computing ability, supercomputing ability, and intelligent computing ability. The commonly used measurement unit is the number of floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS). The larger the value, the stronger the comprehensive computing ability. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 + CP 智能 + CP 超级 。

[0049] The "carrying capacity" (Network Power, NP) as described in the present invention refers to: the manifestation of the data transmission ability of computing power facilities, which is a comprehensive ability including network architecture, network bandwidth, transmission delay, intelligent management and scheduling, etc., and involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling ability.

[0050] The "storage power" (Storage Power, SP) as described in the present invention refers to: the comprehensive ability of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon, which is a comprehensive indicator for measuring the data storage ability of a data center and includes external storage devices such as storage arrays and server internal storage devices. The commonly used measurement unit for storage capacity is exabyte (EB, 1 EB = 2^60 bytes), the commonly used measurement unit for performance is the number of read / write operations per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB), and the disaster recovery ratio is an important manifestation of security and reliability.

[0051] The "computing power infrastructure" as described in the present invention refers to: a new type of information infrastructure integrating information computing power, network carrying capacity, and data storage capacity, which can realize the centralized computing, storage, transmission, and application of information.

[0052] The "new information infrastructure" described in the present invention refers to: mainly including network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, and satellite Internet, computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, and supercomputing centers, and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0053] The "computing power" described in the present invention includes: general computing power, intelligent computing power, and super computing power.

[0054] The "general computing power" described in the present invention refers to: the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0055] The "intelligent computing power" described in the present invention refers to: for various artificial intelligence innovation applications, a computing platform is deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit), such as natural language processing, machine vision, and so on.

[0056] The "super computing power" described in the present invention refers to: mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and processes extremely complex or data-intensive problems through a dedicated operating system, mainly for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, gene analysis, etc.

[0057] The "intelligent computing center" described in the present invention refers to: a facility that provides the required computing power, data, and algorithms mainly for artificial intelligence applications (such as scenarios like artificial intelligence deep learning model development, model training, and model inference) by using large-scale heterogeneous computing power resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0058] The "intelligent computing center" described in the present invention includes but is not limited to the "intelligent computing center".

[0059] The "intelligent computing center" described in the present invention, namely the artificial intelligence computing center, is a type of computing power infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0060] The "computing power center" described in the present invention refers to a facility mainly composed of infrastructure such as wind, fire, water, and electricity, and IT software and hardware devices, which has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0061] The "supercomputing center" described in the present invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters, and can provide functions such as large-scale computing, storage, and network services, and is widely used in application scenarios such as aerospace, national defense, oil exploration, climate modeling, and genome sequencing.

[0062] The "computing power resources" described in the present invention refer to technologies and facilities with information computing, transmission, storage, and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, as well as support and guarantee resources such as wind, fire, water, and electricity.

[0063] The "large language model" described in the present invention refers to a large language model (LLM), which is a language model with a large number of parameters, aiming to understand and generate human language, trained through a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0064] The "Multi-Route Retrieval" described in the present invention refers to a technology in an information retrieval system to improve the comprehensiveness and accuracy of retrieval by obtaining information through multiple different retrieval paths or different retrieval strategies. The core idea of Multi-Route Retrieval is that a single retrieval path may miss some important information or not cover all possible relevant content, so multiple retrieval methods need to be adopted in parallel or serially to improve the final recall rate and accuracy.

[0065] The "Vector Search" described in the present invention is a similarity search technology for efficiently processing large-scale data sets, especially suitable for storing and managing vector data. It converts data into vector representations and stores them in a vector database, and then uses vector similarity algorithms for fast retrieval.

[0066] The "full-text retrieval" described in the present invention refers to performing document morphemes on the user's question and obtaining the retrieval results of computing power resource operation data by calculating the correlation between the morphemes and the documents in the database.

[0067] The "scalar retrieval" described in the present invention refers to the operation of retrieving a single value in a database.

[0068] To solve the problem of difficult query and retrieval of the operation data of computing power resources in an intelligent computing center, please refer to Figure 1 , the present invention provides a multi-way recall retrieval method for the operation data of computing power resources in an intelligent computing center, including:

[0069] Step S1: Obtain a user problem for querying the operation data of computing power resources in the intelligent computing center. The user problem includes at least two of the following retrieval information: vector retrieval information, document retrieval information, and scalar retrieval information;

[0070] Step S2: Analyze the user problem to obtain at least two of the retrieval information in the user problem;

[0071] Step S3: Retrieve the database through the at least two retrieval information respectively to obtain the retrieval results of the operation data of computing power resources that match the at least two retrieval information. The database stores the operation data of multiple computing power resources;

[0072] Step S4: Fit the retrieval results of the operation data of computing power resources that match the at least two retrieval information to obtain the final retrieval results of the operation data of computing power resources.

[0073] In the present invention, by analyzing the input user problem, at least two of the vector retrieval information, document retrieval information, and scalar retrieval information in the user problem are obtained. By querying the database in multiple ways to obtain the retrieval results of the operation data of computing power resources that match each retrieval information, and fitting the retrieval results of the operation data of computing power resources that match each retrieval information, the final retrieval results of the operation data of computing power resources are obtained, solving the problem of difficult query and retrieval of the operation data of computing power resources in the intelligent computing center. Due to the adoption of the multi-way recall retrieval method, the retrieval results of the operation data of computing power resources are more accurate and comprehensive, realizing the precise management of computing power resources.

[0074] In the present invention, at least two types of retrieval information need to be included in the user question. For example, vector retrieval information and document retrieval information can be included simultaneously, or vector retrieval information, document retrieval information, and scalar retrieval information can be included simultaneously. The vector retrieval information is the vector information in the user question, the document retrieval information is the document information in the user question, and the scalar retrieval information is the scalar information in the user question. For example, the user question can be "Please find in the GPU cluster computing power nodes that the real-time temperature monitoring value is greater than 82°C, the running log description shows 'video memory occupancy rate exceeds 90%', and at the same time the load fluctuation of the node in the past 24 hours is close to [0.959139347076416, 0.8759326338768005, 0.35781726241111755, 0.5009950995445251, 0.683462917804718, 0.7127020359039307, 0.37025076150894165, 0.5611962080001831]." Among them, "the temperature monitoring value is greater than 82°C" is scalar retrieval information, "the running log description shows 'video memory occupancy rate exceeds 90%'" is document retrieval information, and "the load fluctuation of the node in the past 24 hours is close to [0.959139347076416, 0.8759326338768005, 0.35781726241111755, 0.5009950995445251, 0.683462917804718, 0.7127020359039307, 0.37025076150894165, 0.5611962080001831]" is vector retrieval information.

[0075] In the present invention, a large language model can be used to analyze the user question. The above user question can be input into the large language model, and the large language model analyzes the user question to obtain each retrieval information in the user question.

[0076] In the present invention, the obtained retrieval information is input into the database, and the database retrieves the stored computing power resource operation data based on each retrieval information.

[0077] In the present invention, in the database, vector data of computing power resources can be stored (for example, time series data of GPU / CPU utilization: recording the GPU / CPU utilization every hour of each day forms a vector, which helps analyze the load situation and trend of computing power resources. Another example is the change trend of memory usage: representing the fluctuations in memory usage with a vector helps understand the memory requirements during system operation. Another example is network traffic patterns: recording the data volume of network inflows and outflows at time intervals and forming vectors can be used to monitor network activity patterns and detect anomalies), scalar data (such as the running data itself), and document data (such as description information of the running data, etc.). Of course, for a certain computing power resource in the database, not all of the above three types of data are required. For example, a computing power resource may only have vector data and scalar data, without document data.).

[0078] In the present invention, optionally, the database can be a multi-modal vector database, which supports scalar, vector, and full-text index retrieval simultaneously.

[0079] The retrieval process of each retrieval information will be described below.

[0080] In the present invention, optionally, the step S3 includes:

[0081] Step S31: If the retrieval information includes vector retrieval information, convert the vector retrieval information into a target vector, calculate the similarity between the target vector and the vector data of each computing power resource in the database, and obtain a first retrieval result according to the calculation result. The first retrieval result includes the computing power resources corresponding to the top N vector data with the highest similarity;

[0082] That is, sort the similarities of the vector data of each computing power resource (Sort) to obtain the computing power resources corresponding to the top N vector data with the highest similarity.

[0083] In the present invention, N is an integer greater than or equal to 1. The value of N can be set in advance.

[0084] It should be noted that in some similarity calculation algorithms, the calculated similarity may be negative, so it is necessary to normalize the calculated similarity.

[0085] In the present invention, the first retrieval result can be shown in Table 1, for example:

[0086]

[0087] Table 1

[0088] Step S32: If the retrieved information includes document retrieval information, decompose the document retrieval information into multiple morphemes, calculate the correlation scores of each morpheme with the document data of each computing power resource in the database, and determine the total correlation score of the document data according to the correlation scores of all morphemes with the document data. Obtain a second retrieval result according to the total correlation score, where the second retrieval result includes the computing power resources corresponding to the top M document data with the highest total correlation scores;

[0089] That is, sort the correlation scores of the document data of each computing power resource to obtain the computing power resources corresponding to the top M document data with the highest total correlation scores.

[0090] In the present invention, M is an integer greater than or equal to 1. The value of M can be the same as or different from the value of N. The value of M can be set in advance.

[0091] For example, assume that the user's question includes: retrieving a GPU with a hardware error log. At this time, the following morphemes can be obtained: CPU and hardware error log.

[0092] Perform a correlation match between each morpheme and the document data of each computing power resource in the database to obtain the correlation scores of the morpheme with the document data of each computing power resource. For the document data of the same computing power resource, calculate the weighted sum of the correlation scores of all morphemes with this document data as the total correlation score of the document data. For example, the correlation score of the document data of a computing power resource with CPU is 90%, and the correlation score with the hardware error log is 0%. Then the total correlation score of this document data can be: 90% * 0.5 + 0% * 0.5 = 45%.

[0093] In the present invention, the weight occupied by each morpheme can be fixed, such as evenly distributing the weights, or can be obtained after analyzing the importance of the morpheme. In the present invention, a large language model can be used to analyze each morpheme to obtain the corresponding weight.

[0094] In the present invention, the correlation scores of each morpheme with the document data of each computing power resource in the database can be calculated according to the BM25 algorithm. Of course, it is not excluded to use other algorithms, such as the TF-IDF (Term Frequency Inverse Document Frequency) algorithm, etc.

[0095] In the present invention, the second retrieval result can be shown in Table 2, for example:

[0096]

[0097] Table 2

[0098] Step S33: If the retrieved information includes scalar retrieval information, query the database to obtain a third retrieval result, where the third retrieval result includes computing power resources that match the scalar retrieval information.

[0099] For example, assume that the user's question includes: The retrieval process occupies more than 90% of the GPUs. Among them, 90% is the scalar retrieval information.

[0100] In the present invention, the third retrieval result can be, for example, as shown in Table 3:

[0101]

[0102] Table 3

[0103] In the present invention, the above-mentioned step S31, step S32, and step S33 can be executed in parallel or serially.

[0104] The following describes the fitting process of the retrieval results of the computing power resource operation data.

[0105] In the present invention, optionally, the step S4 includes:

[0106] Step S41: If the retrieval result of the computing power resource operation data only includes the first retrieval result and the second retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the second retrieval result, and perform a weighted sum of the similarity and relevance total scores of the computing power resources with the same identifier to obtain the top X computing power resources with the highest weighted sum score as the final retrieval result of the computing power resource operation data;

[0107] In the present invention, X is an integer greater than or equal to 1. The value of X can be set in advance.

[0108] Of course, in special cases, if the number of computing power resources with the same identifier is less than or equal to X, the computing power resources with the same identifier can be directly obtained as the final retrieval result of the computing power resource operation data.

[0109] For example, for the retrieval results of the computing power resource operation data in Table 1 and Table 2 above, it can be seen that ID1010078 and ID0050221 are the computing power resources with the same identifier in the two tables. Perform a weighted sum of the similarity and relevance total scores of these computing power resources with the same identifier, and then sort the computing power resources to obtain the top X computing power resources with the highest weighted sum score as the final retrieval result of the computing power resource operation data.

[0110] Step S42: If the retrieved results of the computing power resource operation data include a first retrieved result, a second retrieved result, and a third retrieved result, obtain the computing power resources with the same identifier among the first retrieved result, the second retrieved result, and the third retrieved result, and perform a weighted sum of the similarity and the total relevance score of the computing power resources with the same identifier to obtain the top Y computing power resources with the highest weighted sum score as the final retrieved results of the computing power resource operation data;

[0111] Step S43: If the retrieved results of the computing power resource operation data include a first retrieved result and a third retrieved result, obtain the computing power resources with the same identifier among the first retrieved result and the third retrieved result, and obtain the top Z computing power resources with the highest similarity among the computing power resources with the same identifier as the final retrieved results of the computing power resource operation data;

[0112] Step S44: If the retrieved results of the computing power resource operation data include a second retrieved result and a third retrieved result, obtain the computing power resources with the same identifier among the second retrieved result and the third retrieved result, and obtain the top S computing power resources with the highest total relevance score among the computing power resources with the same identifier as the final retrieved results of the computing power resource operation data.

[0113] The above X, Y, Z, and S are all integers greater than or equal to 1. X, Y, Z, and S can be the same or different.

[0114] In the present invention, optionally, before step S41 or step S42, the following is further included:

[0115] Step S40: Analyze the user question to obtain the weights of the similarity and the total relevance score; or, obtain the pre-set weights of the similarity and the total relevance score.

[0116] That is to say, the above weights can be fixed (pre-set) or variable, obtained by analyzing the user question.

[0117] In the present invention, a large language model can be used to analyze the user question to obtain the weights of the similarity and the total relevance score. For example, it can be analyzed which of the vector retrieval information and the document retrieval information in the user question is more important, so as to obtain the corresponding weights.

[0118] Please refer to Figure 2 , Figure 2 which is a schematic diagram of the intelligent computing center computing power resource operation data multi-way recall retrieval method according to an embodiment of the present invention. In this embodiment, the user question includes: vector retrieval information and document retrieval information.

[0119] In one retrieval path: convert the vector retrieval information in the user's question into a target vector, and perform vector retrieval (vector_search) in the database using the target vector to obtain the similarity between the vector data of each computing power resource in the database and the target vector. Then, normalize the similarities of the vector data of all computing power resources and perform sorting (sort) to obtain the top N computing power resources with the highest similarity.

[0120] In another retrieval path: perform word segmentation on the document retrieval information in the user's question, decompose the document retrieval information into multiple morphemes (queryStr), and perform document retrieval (document_search) in the database using the multiple morphemes (queryStr) to obtain the BM25 scores of each morpheme and the document data of each computing power resource in the database. For a piece of document data, calculate the weighted sum of the BM25 scores of all morphemes and it to obtain the total BM25 score of the document data. Then, sort the total BM25 scores of all document data (sort) to obtain the top M computing power resources with the highest total BM25 score.

[0121] Fit (full join) the two paths of data together based on the key value (identifier) of the computing power resource, that is, obtain the computing power resources with key values in topN and topM, and perform a weighted sum of the similarity and the total BM25 score of the computing power resources with key values to obtain the top X computing power resources with the highest weighted sum score (score_hybrid) as the final retrieval result of the computing power resource operation data (project).

[0122] In the present invention, optionally, the multi-path recall retrieval method for the computing power resource operation data of the intelligent computing center further includes:

[0123] Step S5: Monitor the computing power resource operation data of the intelligent computing center in real time;

[0124] Step S6: Record the monitored computing power resource operation data into the database.

[0125] In the present invention, optionally, the computing power resource includes at least one of the following: computing resource, network resource, storage resource, security resource, and support guarantee resource. Further optionally, the computing resource includes at least one of the following: GPU, CPU, the network resource includes at least one of the following: switch, router. The storage resource includes at least one of the following: storage array, distributed storage. The security resource includes at least one of the following: firewall, intrusion detection system. The support guarantee resource includes at least one of the following: wind power support guarantee resource, thermal power support guarantee resource, hydraulic power support guarantee resource, power support guarantee resource.

[0126] In the present invention, optionally, the computing power resource operation data includes at least one of the following: basic performance index data, power consumption and power management data, video memory related data, heat dissipation and fan control data, hardware status and error monitoring data, process and task management data, performance optimization index data, interface and data transmission data, driver and system information, advanced function monitoring data, tool and scenario application data.

[0127] Taking the GPU as an example below, the computing power resource operation data will be described.

[0128] The operation data of the GPU covers multiple dimensions and is used to monitor performance, diagnose problems and optimize the system. The following are its main classifications and detailed contents:

[0129] 1. Basic performance indicators

[0130] Temperature data: core temperature, video memory temperature, hotspot temperature, to prevent performance degradation or hardware damage caused by overheating.

[0131] Utilization rate: core utilization rate (proportion of tasks processed), video memory utilization rate (proportion of video memory occupied).

[0132] Clock frequency: core frequency, video memory frequency, boost frequency (such as Boost Clock), reflecting the current computing speed.

[0133] 2. Power consumption and power management

[0134] Real-time power consumption: current power consumption and power consumption upper limit (TDP).

[0135] Voltage data: core voltage, video memory voltage, affecting overclocking stability.

[0136] 3. Video memory related data

[0137] Capacity and usage: total video memory, used video memory, to determine whether it becomes a performance bottleneck.

[0138] Bandwidth utilization rate: video memory read and write speed, affecting data processing efficiency.

[0139] 4. Heat dissipation and fan control

[0140] Fan speed: current speed, target speed and temperature-speed curve, to ensure heat dissipation efficiency.

[0141] Heat dissipation events: such as the number of thermal throttling times, reflecting the frequency reduction caused by overheating.

[0142] 5. Hardware status and error detection

[0143] Error Correction (ECC): Error counting in the video memory of professional GPUs to ensure data integrity.

[0144] Hardware Error Log: Records such as driver crashes or hardware failures.

[0145] 6. Process and Task Management

[0146] Process Occupancy: GPU resource usage of each application or process (such as CUDA processes).

[0147] Load Type: 3D rendering, video decoding, computing tasks, etc., to distinguish working modes.

[0148] 7. Performance Optimization Metrics

[0149] Frames Per Second (FPS): Real-time rendering performance, especially crucial in games.

[0150] Utilization of Computing Units: Usage of CUDA cores, Tensor Cores, or ray tracing cores to optimize Artificial Intelligence (AI) or rendering tasks.

[0151] 8. Interfaces and Data Transfer

[0152] Peripheral Component Interconnect Express (PCIe) Bandwidth: Data transfer rate between the GPU and the CPU, affecting the efficiency of large-scale computing tasks.

[0153] Multi-Monitor Output: Resolution, refresh rate, and connection status to optimize multi-screen configurations.

[0154] 9. Driver and System Information

[0155] Driver Version: Affects compatibility and performance, and needs to be updated regularly.

[0156] Firmware Version: Basic Input Output System (BIOS) or firmware status of the GPU to ensure normal functionality.

[0157] 10. Advanced Function Monitoring

[0158] Ray Tracing Data: Ray tracing core load to optimize game or rendering settings.

[0159] AI Acceleration Metrics: Such as Tensor Core utilization to improve the efficiency of deep learning tasks.

[0160] 11. Tools and Scenario Applications

[0161] Monitoring tools: GPU-Z, HWMonitor, NVIDIA SMI, AMD Radeon Software, etc. provide the above data.

[0162] Application scenarios: game optimization, machine learning model training, heat dissipation design verification, etc.

[0163] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of the multi-channel recall retrieval device for the operation data of computing power resources in the intelligent computing center of the present invention. The multi-channel recall retrieval device 10 for the operation data of computing power resources in the intelligent computing center includes:

[0164] An acquisition module 11, configured to acquire a user question, where the user question is used to query the operation data of computing power resources in the intelligent computing center, and the user question includes at least two of the following retrieval information: vector retrieval information, document retrieval information, scalar retrieval information;

[0165] An analysis module 12, configured to analyze the user question to obtain at least two pieces of retrieval information in the user question;

[0166] A retrieval module 13, configured to retrieve the database through the at least two pieces of retrieval information respectively to obtain a retrieval result of the operation data of computing power resources that matches the at least two pieces of retrieval information, where the database stores the operation data of multiple computing power resources;

[0167] A fitting module 14, configured to fit the retrieval result of the operation data of computing power resources that matches the at least two pieces of retrieval information to obtain a final retrieval result of the operation data of computing power resources.

[0168] Optionally, the retrieval module 13 includes:

[0169] A first retrieval sub-module, configured to, if the retrieval information includes vector retrieval information, convert the vector retrieval information into a target vector, calculate the similarity between the target vector and the vector data of each computing power resource in the database, and obtain a first retrieval result according to the calculation result, where the first retrieval result includes the computing power resources corresponding to the top N vector data with the highest similarity;

[0170] A second retrieval sub-module, configured to, if the retrieval information includes document retrieval information, decompose the document retrieval information into multiple morphemes, calculate the correlation score between each morpheme and the document data of each computing power resource in the database, and determine the total correlation score of the document data according to the correlation scores of all morphemes and the document data, and obtain a second retrieval result according to the total correlation score, where the second retrieval result includes the computing power resources corresponding to the top M document data with the highest total correlation score;

[0171] A third retrieval sub-module, configured to query the database to obtain a third retrieval result if the retrieval information includes scalar retrieval information, where the third retrieval result includes computing power resources that match the scalar retrieval information.

[0172] Optionally, the fitting module 14 includes:

[0173] A first fitting sub-module, configured to, if the computing power resource operation data retrieval result only includes a first retrieval result and a second retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the second retrieval result, and perform a weighted sum of the similarity and the total relevance score of the computing power resources with the same identifier, and obtain the top X computing power resources with the highest weighted sum score as the final computing power resource operation data retrieval result;

[0174] A second fitting sub-module, configured to, if the computing power resource operation data retrieval result includes a first retrieval result, a second retrieval result, and a third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result, the second retrieval result, and the third retrieval result, and perform a weighted sum of the similarity and the total relevance score of the computing power resources with the same identifier, and obtain the top Y computing power resources with the highest weighted sum score as the final computing power resource operation data retrieval result;

[0175] A third fitting sub-module, configured to, if the computing power resource operation data retrieval result includes a first retrieval result and a third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the third retrieval result, and obtain the top Z computing power resources with the highest similarity in the computing power resources with the same identifier as the final computing power resource operation data retrieval result;

[0176] A fourth fitting sub-module, configured to, if the computing power resource operation data retrieval result includes a second retrieval result and a third retrieval result, obtain the computing power resources with the same identifier in the second retrieval result and the third retrieval result, and obtain the top S computing power resources with the highest total relevance score in the computing power resources with the same identifier as the final computing power resource operation data retrieval result.

[0177] Optionally, the intelligent computing center computing power resource operation data multi-way recall retrieval device 10 further includes:

[0178] A weight acquisition module, configured to analyze the user's question to obtain the weights of the similarity and the total relevance score; or, obtain the pre-set weights of the similarity and the total relevance score.

[0179] Optionally, the intelligent computing center computing power resource operation data multi-way recall retrieval device 10 further includes:

[0180] A monitoring module, configured to monitor in real time the operation data of the computing power resources of the intelligent computing center;

[0181] A storage module, configured to record the monitored operation data of the computing power resources into the database.

[0182] Optionally, the computing power resources include at least one of the following: computing resources, network resources, storage resources, security resources, and support and guarantee resources;

[0183] The operation data of the computing power resources includes at least one of the following: basic performance index data, power consumption and power management data, video memory related data, heat dissipation and fan control data, hardware status and error monitoring data, process and task management data, performance optimization index data, interface and data transmission data, driver and system information, advanced function monitoring data, tool and scenario application data.

[0184] Please refer to Figure 4 , the present invention further provides an electronic device 20, including a processor 21, a memory 22, and a computer program stored on the memory 22 and executable on the processor 21. When the computer program is executed by the processor 21, it implements each process of the above-mentioned embodiment of the method for multi-way recall retrieval of the operation data of the computing power resources of the intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0185] The present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements each process of the above-mentioned embodiment of the method for multi-way recall retrieval of the operation data of the computing power resources of the intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0186] The present invention further provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement each process of the above-mentioned Figure 1 method embodiment shown, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0187] It should be noted that in this text, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device comprising that element.

[0188] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal (which can be a mobile device, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0189] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them fall within the protection scope of the present invention.

Claims

1. A method for retrieving multiple runs of computing power resource operation data in an intelligent computing center, characterized in that, Including: Step S1: Obtain a user question for querying the computing power resource operation data of an intelligent computing center. The user question includes at least two of the following retrieval information: vector retrieval information, document retrieval information, and scalar retrieval information; Step S2: Analyze the user question to obtain at least two types of retrieval information in the user question; Step S3: Retrieve the database respectively through the at least two types of retrieval information to obtain a retrieval result of computing power resource operation data that matches the at least two types of retrieval information. The operation data of multiple computing power resources is stored in the database; Step S4: Fit the retrieval result of computing power resource operation data that matches the at least two types of retrieval information to obtain a final retrieval result of computing power resource operation data.

2. The method according to claim 1, characterized in that The step S3 includes: Step S31: If the retrieval information includes vector retrieval information, convert the vector retrieval information into a target vector, calculate the similarity between the target vector and the vector data of each computing power resource in the database, and obtain a first retrieval result according to the calculation result. The first retrieval result includes the computing power resources corresponding to the top N vector data with the highest similarity; Step S32: If the retrieval information includes document retrieval information, decompose the document retrieval information into multiple morphemes, calculate the correlation score between each morpheme and the document data of each computing power resource in the database, and determine the total correlation score of the document data according to the correlation scores of all morphemes and the document data. Obtain a second retrieval result according to the total correlation score. The second retrieval result includes the computing power resources corresponding to the top M document data with the highest total correlation score; Step S33: If the retrieval information includes scalar retrieval information, query the database to obtain a third retrieval result. The third retrieval result includes the computing power resources that match the scalar retrieval information.

3. The method according to claim 2, wherein The step S4 includes: Step S41: If the retrieval result of computing power resource operation data only includes the first retrieval result and the second retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the second retrieval result, and perform a weighted sum of the similarity and the total correlation score of the computing power resources with the same identifier to obtain the top X computing power resources with the highest weighted sum score as the final retrieval result of computing power resource operation data; Step S42: If the retrieval result of computing power resource operation data includes the first retrieval result, the second retrieval result, and the third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result, the second retrieval result, and the third retrieval result, and perform a weighted sum of the similarity and the total correlation score of the computing power resources with the same identifier to obtain the top Y computing power resources with the highest weighted sum score as the final retrieval result of computing power resource operation data; Step S43: If the computing power resource operation data retrieval result includes a first retrieval result and a third retrieval result, obtain the computing power resources with the same identifier in the first retrieval result and the third retrieval result, and obtain the top Z computing power resources with the highest similarity in the computing power resources with the same identifier as the final computing power resource operation data retrieval result; Step S44: If the computing power resource operation data retrieval result includes a second retrieval result and a third retrieval result, obtain the computing power resources with the same identifier in the second retrieval result and the third retrieval result, and obtain the top S computing power resources with the highest total relevance score in the computing power resources with the same identifier as the final computing power resource operation data retrieval result.

4. The method according to claim 3, wherein Before step S41 or step S42, it further includes: Step S40: Analyze the user question to obtain the weights of the similarity and the total relevance score; or, obtain the pre-set weights of the similarity and the total relevance score.

5. The method according to claim 1, wherein It further includes: Step S5: Real-time monitor the computing power resource operation data of the intelligent computing center; Step S6: Record the monitored computing power resource operation data into the database.

6. The method according to any one of claims 1-5, characterized in that The computing power resources include at least one of the following: computing resources, network resources, storage resources, security resources, and support and guarantee resources; The computing power resource operation data includes at least one of the following: basic performance index data, power consumption and power management data, video memory related data, heat dissipation and fan control data, hardware status and error monitoring data, process and task management data, performance optimization index data, interface and data transmission data, driver and system information, advanced function monitoring data, tool and scenario application data.

7. An intelligent computing center computing power resource operation data multi-channel recall and retrieval device, characterized in that, It includes: An acquisition module, configured to acquire a user question, where the user question is used to query the computing power resource operation data of the intelligent computing center, and the user question includes at least two of the following retrieval information: vector retrieval information, document retrieval information, scalar retrieval information; An analysis module, configured to analyze the user question to obtain at least two of the retrieval information in the user question; A retrieval module, configured to retrieve the database respectively through the at least two retrieval information to obtain a computing power resource operation data retrieval result matching the at least two retrieval information, and the database stores the operation data of multiple computing power resources; A fitting module, configured to fit the computing power resource operation data retrieval result matching the at least two retrieval information to obtain the final computing power resource operation data retrieval result.

8. An electronic device, characterized in that, It includes: A processor, a memory, and a program stored on the memory and executable on the processor, where the program, when executed by the processor, implements the steps of the method for multi-way recall retrieval of computing power resource operation data of an intelligent computing center according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the method for multi-way recall retrieval of computing power resource operation data of an intelligent computing center according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Including computer instructions, when the computer instructions are executed by a processor, the steps of the method for retrieving and recalling operation data of computing power resources of an intelligent computing center as described in any one of claims 1 to 6 are implemented.

Images