Rule determination method and device for realizing computing power of intelligent computing center

Abstract

The invention provides a rule determination method and device for computing power implementation of an intelligent computing center. The method comprises the steps that S1, a task to be computed is received; s2, in response to the to-be-calculated task, extracting problem characteristics representing reasoning task characteristics from the to-be-calculated task; furthermore, obtaining rule information of each to-be-selected rule in a preset to-be-selected rule set, and calling a feature extraction large language model deployed in an intelligent computing center to reasone the rule information to obtain rule features of each to-be-selected rule; s3, vectorizing the problem features to obtain problem feature vectors; vectorizing the rule features to obtain rule feature vectors of the to-be-selected rules; and S4, according to the cosine similarity between the problem feature vector and each rule feature vector, determining a target rule from the to-be-selected rule set. According to the invention, the target rule can be determined with high accuracy and high efficiency.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure, and particularly relates to a method and device for determining rules for realizing computing power 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, and mainly provides 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). An intelligent computing center encompasses 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 that is based on artificial intelligence theory, adopts an artificial intelligence computing architecture, and provides 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 a computer device or a computing / data center to process information. It is the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement. It is the computing ability to output a target result by processing information data. It is a new type of productive force that integrates information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0007] When performing a computing task, it is necessary to determine the rules for executing the computing task. Existing rule determination methods only rely on the problem characteristics of the task to be computed, without considering the matching of the rules with the task to be computed, resulting in low accuracy in determining rules. Moreover, most of them are manually determined by analyzing the problem characteristics of the task to be computed, with low efficiency in determining rules. Summary of the Invention

[0008] The present invention provides a method and device for determining rules for realizing computing power in an intelligent computing center to solve the problem that existing rule determination methods only rely on the problem characteristics of the task to be computed, without considering the matching of the rules with the task to be computed, resulting in low accuracy in determining rules. Moreover, most of them are manually determined by analyzing the problem characteristics of the task to be computed, with low efficiency in determining rules.

[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 determining rules for realizing computing power in an intelligent computing center, including:

[0011] Step S1: Receive a task to be calculated;

[0012] Step S2: In response to the task to be calculated, extract problem features characterizing the characteristics of the inference task from the task to be calculated; and, obtain the rule information of each candidate rule in a preset candidate rule set, and call a feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule;

[0013] Step S3: Vectorize the problem features to obtain a problem feature vector; vectorize the rule features to obtain rule feature vectors of each candidate rule;

[0014] Step S4: Determine a target rule for executing the task to be calculated from the candidate rule set according to the cosine similarity between the problem feature vector and each rule feature vector.

[0015] Optionally, the feature extraction large language model includes: a code-specific feature extraction large language model, a general large language model;

[0016] Calling a feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule includes:

[0017] Step S21: Judge the data type of the rule information to obtain a judgment result;

[0018] Step S22: If the judgment result indicates that the rule information includes information in the form of a rule code, input the information in the form of a rule code into the code-specific feature extraction large language model to obtain a first rule feature of the information in the form of a rule code;

[0019] Step S23: If the judgment result indicates that the rule information includes information in the form of a text description, generate a prompt word for instructing the large language model to obtain features from the information in the form of a text description, and input the prompt word and the information in the form of a text description into the general large language model to obtain a second rule feature of the information in the form of a text description;

[0020] Step S24: Combine the first rule feature and the second rule feature to obtain the rule feature.

[0021] Optionally, the code-specific feature extraction large language model includes at least one of the following models: CodeBERT model, UniXCoder model; and / or,

[0022] The general large language model includes at least one of the following models: DeepSeek model, ChatGLM model.

[0023] Optionally, the rule information includes at least one of the following:

[0024] Programming language code, abstract syntax tree (AST) of the code, macro statistical features of the code, pseudocode, papers, manuals, hyperparameters of the rules, execution process of the rules, model structure of the rules, loss function of the rules.

[0025] Optionally, the problem features include at least one of the following:

[0026] Scale of the input data required to execute the to-be-computed task, data type of the input data required to execute the to-be-computed task, attributes and constraint conditions of the problem to be solved by the to-be-computed task.

[0027] Optionally, step S4 includes:

[0028] Step S41: For each of the rule feature vectors, concatenate the problem feature vector, the rule feature vector, and the corresponding cosine similarity to obtain a candidate combination;

[0029] Step S42: Input all the candidate combinations into a multi-layer perceptron (MLP) model deployed in the intelligent computing center to obtain the matching degree between the to-be-computed task and each of the candidate rules, and obtain the predicted execution effect after the to-be-computed task is executed according to each of the candidate rules;

[0030] Step S43: Determine the target rule according to the matching degree and the predicted execution effect.

[0031] In a second aspect, an apparatus for determining rules for realizing computing power in an intelligent computing center according to the present invention includes:

[0032] A receiving module, configured to receive a to-be-computed task;

[0033] An execution module, configured to, in response to the to-be-computed task, extract problem features characterizing the inference task characteristics from the to-be-computed task; and obtain the rule information of each candidate rule in a preset candidate rule set, and call a feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule;

[0034] A vectorization module, configured to vectorize the problem features to obtain a problem feature vector; and vectorize the rule features to obtain the rule feature vectors of each candidate rule;

[0035] A determination module, configured to determine, according to the cosine similarity between the problem feature vector and each of the rule feature vectors, a target rule for performing the to-be-calculated task from the set of candidate rules.

[0036] In a third aspect, an electronic device according to the present invention includes a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor, the steps in the rule determination method for realizing the computing power of the intelligent computing center according to any one of the first aspects are implemented.

[0037] In a fourth aspect, a readable storage medium according to the present invention stores a program or instruction thereon. When the program or instruction is executed by a processor, the steps in the rule determination method for realizing the computing power of the intelligent computing center according to any one of the first aspects are implemented.

[0038] In a fifth aspect, a computer program product according to the present invention includes computer instructions. When the computer instructions are executed by a processor, the steps of the rule determination method for realizing the computing power of the intelligent computing center according to any one of the first aspects are implemented.

[0039] In the present invention, through step S1: receiving a to-be-calculated task; step S2: in response to the to-be-calculated task, extracting problem features characterizing the inference task characteristics from the to-be-calculated task; and obtaining the rule information of each candidate rule in the preset set of candidate rules, and invoking the feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule; step S3: vectorizing the problem features to obtain a problem feature vector; vectorizing the rule features to obtain the rule feature vectors of each candidate rule; step S4: according to the cosine similarity between the problem feature vector and each rule feature vector, determining a target rule for performing the to-be-calculated task from the set of candidate rules. The present invention realizes the determination of the target rule based on the problem features of the to-be-calculated task and the rule features of the candidate rules, and can accurately determine the target rule; in addition, the present invention uses the feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule, and applies the abundant computing power of the intelligent computing center to improve the efficiency of determining the rule features, and can efficiently determine the target rule. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By reading the following detailed description of the preferred embodiments, 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 1Schematic flowchart of the method for determining the rules for realizing the computing power of the intelligent computing center of the present invention;

[0042] Figure 2 Principle block diagram of the device for determining the rules for realizing the computing power of the intelligent computing center of the present invention;

[0043] Figure 3 Principle block diagram of the electronic device of the present invention. Detailed implementation manners

[0044] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0045] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances so that the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are usually of the same type, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "or" in the present invention means at least one of the connected objects. For example, "A or B" covers three scenarios, namely, Scenario 1: including A and not including B; Scenario 2: including B and not including A; Scenario 3: including both A and B. The character " / " generally indicates an "or" relationship between the associated objects before and after.

[0046] In the technical solutions of the present invention, terms such as "connected", "coupled" or "linked" do not limit to physical or mechanical connections, but may include electrical connections.

[0047] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0048] It should be noted that in the technical solutions of the present invention, aspects such as the collection, collection, update, analysis, processing, use, transmission, and storage of personal information all comply with the provisions of relevant laws and regulations, are used for legal purposes, and do not violate public order and good customs. Necessary measures are taken for personal information to prevent illegal access to personal information data and to maintain personal information security and network security.

[0049] Next, the technical terms involved in the present invention will be briefly described first.

[0050] The "computing power" 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 execute a certain computing requirement together, the computing ability to achieve the output of a target result by processing information data, and 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.

[0051] The "computational power" (CP) 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 to measure 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), and 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 .

[0052] The "network power" (NP) 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 inside and between data centers, and is a comprehensive indicator to measure network transmission scheduling ability.

[0053] The "storage power" (SP) 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 to measure 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 and 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.

[0054] The "computing power infrastructure" 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, and presents characteristics such as multi-source ubiquitous, intelligent and agile, secure and reliable, and green and low-carbon, and is of great significance for boosting industrial transformation and upgrading, empowering China's scientific and technological innovation, meeting people's beautiful life, and realizing high-efficiency social governance.

[0055] 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, satellite Internet, etc., computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, supercomputing centers, etc., and new technology facilities such as artificial intelligence, blockchain, and quantum computing. With the emergence and popularization of new general technologies, the form of the new information infrastructure will be more diverse.

[0056] The "computing power" described in the present invention includes: general computing power, intelligent computing power, and super computing power.

[0057] 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.

[0058] The "intelligent computing power" described in the present invention refers to: for various artificial intelligence innovation applications, a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit), etc., such as natural language processing, machine vision, and so on.

[0059] 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, and is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, gene analysis, etc.

[0060] 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 the underlying computing power to the top-level application enabling.

[0061] The "intelligent computing center" described in the present invention includes but is not limited to the "intelligent computing center".

[0062] The "intelligent computing center" described in the present invention, namely the artificial intelligence computing center, is a type of computing 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.

[0063] 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, with computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0064] The "supercomputing center" described in the present invention refers to: that is, the supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters, capable of providing 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.

[0065] 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, and support and guarantee resources such as wind, fire, water, and electricity.

[0066] The "large language model" described in the present invention refers to the large language model (LLM), which is a language model with a relatively 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.

[0067] The present invention provides a method for determining the rules for realizing the computing power of an intelligent computing center. Refer to Figure 1 as shown in Figure 1 which is a schematic flowchart of the method for determining the rules for realizing the computing power of the intelligent computing center of the present invention, including:

[0068] Step S1: Receive the task to be calculated;

[0069] Step S2: In response to the task to be calculated, extract the problem features characterizing the inference task characteristics from the task to be calculated; and, obtain the rule information of each candidate rule in the preset candidate rule set, and call the feature extraction large language model inference rule information deployed in the intelligent computing center to obtain the rule features of each candidate rule;

[0070] Step S3: Vectorize the problem features to obtain the problem feature vector; vectorize the rule features to obtain the rule feature vectors of each candidate rule;

[0071] Step S4: Determine the target rule for performing the task to be calculated from the set of candidate rules according to the cosine similarity between the problem feature vector and each rule feature vector.

[0072] The rule in the present invention can be an algorithm for calculating the task to be calculated.

[0073] In step S1 of the present invention, the task to be calculated can be sent by the user through the interaction terminal; it can also be a calculation task generated during the operation of the software system itself. The software system sends the calculation task to the execution entity of the present invention, which then becomes the task to be calculated.

[0074] It should be noted that the rule feature is the structural and semantic information extracted from the rule code or description (i.e., the rule information in the present invention), including key characteristics such as the hyperparameters of the rule, the model structure, and the execution process, and is used to judge the applicability of the rule in a specific task. In some optional embodiments, the rule information includes at least one of the following: programming language code, the abstract syntax tree (AST) of the code, the macro statistical features of the code, pseudocode, papers, manuals, the hyperparameters of the rule, the execution process of the rule, the model structure of the rule, and the loss function of the rule. By invoking the feature extraction large language model deployed in the intelligent computing center to infer the above types of rule information, the rule features that accurately represent the specific characteristics of the rule can be obtained.

[0075] The problem feature is the characteristic information describing the inference task (i.e., characterizing the characteristics of the inference task). For example: the scale of the input data, the data type, and the specific attributes of the problem. The problem feature is used to select a suitable rule (i.e., the target rule). In some optional embodiments, the problem feature includes at least one of the following: the scale of the input data required to perform the task to be calculated, the data type of the input data required to perform the task to be calculated, the attributes and constraint conditions of the problem to be solved by the task to be calculated.

[0076] In some embodiments of the present invention, optionally, the feature extraction large language model is a pre-trained large language model. Through pre-training, the feature extraction large language model has the ability to extract rule features from rule information. The relevant pre-training methods are pre-training methods in the prior art and are not the improvement points of the present invention, so they will not be elaborated here. It should be noted that not elaborating should not be considered unclear.

[0077] In the present invention, step S3 is a vectorization step. Through vectorization, the problem features and rule features can be respectively mapped into a high-dimensional continuous vector space to obtain a problem feature vector and a rule feature vector, so as to facilitate step S4 to measure the suitability between the problem and the rule through cosine similarity. Specific methods of vectorization, for example: using LLM-Embedding for vectorization. LLM-Embedding refers to the process of converting words, sentences, paragraphs or even the entire document into real-number vectors, so that machine learning models can understand and process this language data. Embedding is a mapping that can map discrete symbols (such as words, images, documents) into a continuous vector space, enabling the model to perform mathematical operations and processing. Specific methods of vectorization, for another example: using LSTM (Long Short-Term Memory Neural Network) for vectorization. Specifically, when using LSTM for vectorization: considering the sequentiality of the rule features, use LSTM to encode the vector sequence generated by the LLM to obtain the rule feature vector.

[0078] In some alternative embodiments, step S3 includes: inputting the problem features into a multi-layer perceptron (MLP) model to generate a problem feature vector; inputting the rule features into another multi-layer perceptron (MLP) model to generate a rule feature vector. A multi-layer perceptron (MLP) is a neural network structure that realizes the transformation and mapping of the input feature vector through the connection of multiple layers of neurons. In the present invention, the MLP is used to map the rule features and problem features into a unified feature vector.

[0079] In practical applications, there are also problems of too high dimensionality and complexity of the rule feature vector. Based on this, in some alternative embodiments, a feature selection model deployed in the intelligent computing center can be called to screen each feature point in the rule feature vector, and the feature points associated with the rule selection task (that is, the task of performing steps S1 to S4 to realize the determination of the target rule) are taken as key feature points, and the feature vector composed of the key feature points is used as the new rule feature vector for subsequent step S4. Here, "associated" can be understood as: directly related to the effectiveness of the rule selection task (for example: associated inference speed, accuracy).

[0080] Cosine Similarity is a metric used to measure the similarity between two non-zero vectors, and is commonly used in fields such as text analysis, information retrieval, and recommendation systems. It evaluates the similarity by calculating the cosine value of the two vectors, and the value ranges from -1 to 1. In the present invention, cosine similarity is used to measure the similarity between the rule feature vector and the problem feature vector, indicating the matching degree between the two. The higher the similarity, the more matching the rule is with the problem. By using cosine similarity in step S4 to measure the matching degree between the task to be calculated and each candidate rule, and determining the target rule based on the matching degree, a rule with a high matching degree with the task to be calculated can be selected, improving the execution efficiency of the task to be calculated; moreover, since the rule has a high matching degree with the task to be calculated, it is beneficial to reduce the waste of computing power resources when executing the task to be calculated and improve the utilization efficiency of computing power resources.

[0081] In the present invention, through step S1: receiving the task to be calculated; step S2: in response to the task to be calculated, extracting the problem features characterizing the inference task characteristics from the task to be calculated; and obtaining the rule information of each candidate rule in the preset candidate rule set, and invoking the feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule; step S3: vectorizing the problem features to obtain a problem feature vector; vectorizing the rule features to obtain the rule feature vectors of each candidate rule; step S4: determining the target rule for executing the task to be calculated from the candidate rule set according to the cosine similarity between the problem feature vector and each rule feature vector. The present invention realizes the determination of the target rule based on the problem features of the task to be calculated and the rule features of the candidate rules, and can accurately determine the target rule; in addition, the present invention uses the feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule, and applies the abundant computing power of the intelligent computing center to improve the efficiency of determining the rule features, and can efficiently determine the target rule.

[0082] In addition, the present invention uses the feature extraction large language model to extract rule features, which can adapt to different types of rules and support the rapid deployment of new rules. Whether it is a newly added rule or an updated inference task, the present invention can achieve efficient adaptation through retraining and has good scalability.

[0083] In some embodiments of the present invention, optionally, the feature extraction large language model includes: a code-specific feature extraction large language model, a general large language model;

[0084] Invoking the feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule includes:

[0085] Step S21: Determine the data type of the rule information to obtain a judgment result;

[0086] Step S22: If the judgment result indicates that the rule information includes information in the form of a rule code, input the information in the form of a rule code into the large language model for extracting special features of the code to obtain the first rule feature of the information in the form of a rule code;

[0087] Step S23: If the judgment result indicates that the rule information includes information in the form of a text description, generate a prompt word for instructing the large language model to obtain features from the information in the form of a text description, and input the prompt word and the information in the form of a text description into the general large language model to obtain the second rule feature of the information in the form of a text description;

[0088] Step S24: Combine the first rule feature and the second rule feature to obtain the rule feature.

[0089] It should be noted that the judgment result obtained in Step S21 includes: the rule information only includes information in the form of a rule code, the rule information only includes information in the form of a text description, and the rule information includes information in the form of a rule code and information in the form of a text description.

[0090] It can be understood that for the case where the rule information includes information in the form of a rule code and information in the form of a text description, Step S24 can obtain the rule feature by combining the first rule feature and the second rule feature.

[0091] For the case where the rule information only includes information in the form of a rule code, Step S23 is not executed. When executing Step S24, the second rule feature is regarded as blank. That is, in Step S24, the rule feature is obtained by combining the first rule feature and the blank feature. That is to say, the first rule feature is the rule feature. For the case where the rule information only includes information in the form of a text description, Step S22 is not executed. When executing Step S24, the first rule feature is regarded as blank. That is, in Step S24, the rule feature is obtained by combining the blank feature and the second rule feature. That is to say, the second rule feature is the rule feature.

[0092] For the execution process of Step S23, an exemplary scenario is used for illustration. For example: descriptive text (i.e., information in the form of a text description): "Use gradient descent to optimize the loss function, learning rate 0.01, iterate 100 times". Corresponding prompt word: "Extract rule logic, optimizer type, and hyperparameters". Input the prompt word and the information in the form of a text description into the general large language model, and the general large language model outputs the second rule feature including feature points such as "gradient descent" and "learning rate 0.01".

[0093] In the present invention, through steps S21 to S24, a targeted rule feature extraction method is adopted for different forms of information, improving the accuracy of the obtained rule features. Moreover, the two models used (i.e., the large language model for code-specific feature extraction and the general large language model) are both deployed in the intelligent computing center, and the abundant computing power of the intelligent computing center is utilized to improve the execution efficiency of steps S22 and S23, thereby improving the efficiency of determining the target rule in the present invention.

[0094] In some embodiments of the present invention, optionally, the large language model for code-specific feature extraction includes at least one of the following models: CodeBERT model, UniXCoder model; and / or,

[0095] The general large language model includes at least one of the following models: DeepSeek model, ChatGLM model.

[0096] CodeBERT is a pre-trained model based on Transformer, specifically designed for tasks between programming languages and natural languages. It was proposed by Microsoft Research and aims to improve the performance of tasks such as code understanding, generation, and translation. CodeBERT can be used in various applications, including code search, code completion, code translation (such as from one programming language to another), and generation of natural language descriptions of code.

[0097] UniXCoder is a pre-trained model for programming language and natural language understanding, aiming to improve the performance of tasks such as code generation, code translation, and code understanding. It was proposed by Microsoft Research and particularly focuses on cross-language code understanding and generation tasks. The design concept of UniXCoder is to be able to handle multiple programming languages and also handle interactions with natural language.

[0098] DeepSeek is a deep learning model for code understanding and generation, aiming to handle the relationship between code and natural language in a more effective way. This model is mainly used for tasks such as code search, code completion, and code generation, and can help developers program more efficiently.

[0099] ChatGLM is a dialogue model based on the Generative Pretrained Transformer (GPT) architecture, mainly used for natural language processing tasks. It is usually used for generating dialogue content, answering questions, text generation, etc.

[0100] In the present invention, by applying the above models, the accuracy of the rule features can be improved, and thus the accuracy of the determined target rule can be improved.

[0101] In some embodiments of the present invention, optionally, the rule information includes at least one of the following:

[0102] Programming language code, the abstract syntax tree (AST) of the code, the macro statistical features of the code, pseudocode, papers, manuals, hyperparameters of the rule, the execution process of the rule, the model structure of the rule, the loss function of the rule.

[0103] In the present invention, the programming language code, such as: Python, Java, etc. The abstract syntax tree (AST) is an important concept in programming language processing and can effectively represent the structure and semantics of the code. By using the AST, developers and tools can perform various code analysis, optimization, and transformation tasks, thereby improving code quality and development efficiency. The macro statistical features of the code, such as: the number of lines of code, cyclomatic complexity, etc.

[0104] It should be noted that the programming language code, the abstract syntax tree (AST) of the code, and the macro statistical features of the code are information in the form of rule code. The pseudocode, papers, manuals, hyperparameters of the rule, the execution process of the rule, the model structure of the rule, and the loss function of the rule are information in the form of text descriptions.

[0105] Through the above rule information, the present invention can obtain rule features with high accuracy, thereby improving the accuracy of the obtained target rules.

[0106] In some embodiments of the present invention, optionally, the problem features include at least one of the following:

[0107] The scale of the input data required to execute the task to be calculated, the data type of the input data required to execute the task to be calculated, the attributes and constraint conditions of the problem to be solved by the task to be calculated.

[0108] The above problem features can accurately characterize the true characteristics of the task to be calculated. Based on the above problem features, the accuracy of the target rules obtained by the present invention can be improved.

[0109] In some embodiments of the present invention, optionally, step S4 includes:

[0110] Step S41: For each rule feature vector, concatenate the problem feature vector, the rule feature vector, and the corresponding cosine similarity to obtain a candidate combination;

[0111] Step S42: Input all the candidate combinations into the multi-layer perceptron (MLP) model deployed in the intelligent computing center to obtain the matching degree between the task to be calculated and each candidate rule, and obtain the predicted execution effect after the task to be calculated is executed according to each candidate rule;

[0112] Step S43: Determine the target rule according to the matching degree and the predicted execution effect.

[0113] In the present invention, the number of candidate combinations is the same as the number of candidate rules. That is, a problem feature vector is configured for each candidate rule vector, and the cosine similarity between the candidate rule vector and the problem feature vector is configured. That is to say, a candidate combination = problem feature vector + candidate rule feature vector + cosine similarity therebetween. Further, through step S42, all candidate combinations are input into a multi-layer perceptron (MLP) model deployed in the intelligent computing center to obtain the matching degree between the task to be calculated and each candidate rule, and the predicted execution effect after the task to be calculated is executed according to each candidate rule.

[0114] In some alternative embodiments, the predicted execution effect, for example: predicted execution accuracy, predicted execution speed.

[0115] In some embodiments of the present invention, optionally, step S43 is specifically to determine the rule that takes into account both high matching degree and excellent execution effect as the target rule. Step S43 may specifically include: calculating the comprehensive score of each candidate combination, where the comprehensive score of the candidate combination = matching degree × weight of the matching degree + predicted execution effect × weight of the predicted execution effect, and determining the rule in the candidate combination with the highest comprehensive score as the target rule. It can be understood that the user can determine the weights according to their own needs. For example, if the user believes that for the target rule, the matching degree is more important than the execution effect, the weight of the matching degree can be set > the weight of the predicted execution effect; for another example, if the user believes that for the target rule, the execution effect is more important than the matching degree, the weight of the matching degree can be set < the weight of the predicted execution effect.

[0116] In some embodiments of the present invention, optionally, the matching degree can be classified. Specifically, high matching degree: cosine similarity > 0.8 (indicating that the rule highly matches the problem features). Medium matching degree: 0.5 ≤ similarity ≤ 0.8 (indicating that some features of the rule match the problem features). Low matching degree: similarity < 0.5 (indicating that the feature adaptability between the rule and the problem features is poor). Thus, step S43 is specifically to determine the rule that takes into account both high matching degree and excellent execution effect as the target rule, which can be understood as determining the rule with the optimal predicted execution effect from multiple rules with high matching degrees as the target rule.

[0117] In some embodiments of the present invention, optionally, after step S42, linkage optimization of rule selection and computing power allocation can also be performed: high matching degree rules → allocate more computing power (for example: 4 GPUs) to ensure efficient execution of complex tasks. Low matching degree rules → reduce computing power (for example: 1 GPU) to avoid resource waste. Logic: The matching degree reflects the rule adaptability. The higher the adaptability, the more worthy it is to invest resources to improve the execution efficiency.

[0118] The present invention provides a device for determining rules for realizing computing power in an intelligent computing center. SeeFigure 2 As shown Figure 2 Figure 2 is a schematic block diagram of a rule determination device for realizing the computing power of the intelligent computing center of the present invention. The rule determination device 20 for realizing the computing power of the intelligent computing center includes:

[0119] A receiving module 21, configured to receive a task to be calculated;

[0120] An execution module 22, configured to, in response to the task to be calculated, extract problem features characterizing the inference task characteristics from the task to be calculated; and obtain the rule information of each candidate rule in a preset candidate rule set, and call a feature extraction large language model deployed in the intelligent computing center to infer the rule information to obtain the rule features of each candidate rule;

[0121] A vectorization module 23, configured to vectorize the problem features to obtain a problem feature vector; and vectorize the rule features to obtain rule feature vectors of each candidate rule;

[0122] A determination module 24, configured to determine a target rule for executing the task to be calculated from the candidate rule set according to the cosine similarity between the problem feature vector and each rule feature vector.

[0123] In some embodiments of the present invention, optionally, the feature extraction large language model includes: a code-specific feature extraction large language model, a general large language model;

[0124] The execution module 22 is further configured to judge the data type of the rule information to obtain a judgment result;

[0125] The execution module 22 is further configured to, if the judgment result indicates that the rule information includes information in the form of a rule code, input the information in the form of the rule code into the code-specific feature extraction large language model to obtain a first rule feature of the information in the form of the rule code;

[0126] The execution module 22 is further configured to, if the judgment result indicates that the rule information includes information in the form of a text description, generate a prompt word for instructing the large language model to obtain features from the information in the form of the text description, and input the prompt word and the information in the form of the text description into the general large language model to obtain a second rule feature of the information in the form of the text description;

[0127] The execution module 22 is further configured to combine the first rule feature and the second rule feature to obtain the rule feature.

[0128] In some embodiments of the present invention, optionally, the code-specific feature extraction large language model includes at least one of the following models: CodeBERT model, UniXCoder model; and / or,

[0129] The general large language model includes at least one of the following models: Deepseek model, ChatGLM model.

[0130] In some embodiments of the present invention, optionally, the rule information includes at least one of the following:

[0131] Programming language code, abstract syntax tree (AST) of the code, macro statistical features of the code, pseudocode, papers, manuals, hyperparameters of the rules, execution process of the rules, model structure of the rules, loss function of the rules.

[0132] In some embodiments of the present invention, optionally, the problem features include at least one of the following:

[0133] The scale of the input data required to execute the to-be-calculated task, the data type of the input data required to execute the to-be-calculated task, the attributes and constraint conditions of the problem to be solved by the to-be-calculated task.

[0134] In some embodiments of the present invention, optionally, the determining module 24 is further configured to splice the problem feature vector, the rule feature vector, and the corresponding cosine similarity for each of the rule feature vectors to obtain a candidate combination;

[0135] The determining module 24 is further configured to input all the candidate combinations into a multi-layer perceptron (MLP) model deployed in the intelligent computing center to obtain the matching degree between the to-be-calculated task and each of the candidate rules, and to obtain the predicted execution effect after the to-be-calculated task is executed according to each of the candidate rules;

[0136] The determining module 24 is further configured to determine the target rule according to the matching degree and the predicted execution effect.

[0137] The rule determination device for realizing the computing power of the intelligent computing center provided by the present invention can realize Figure 1 Each process implemented by the method embodiment and achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0138] The present invention provides an electronic device 30, as shown in Figure 3 shown. Figure 3FIG. 0 is a schematic block diagram of the electronic device 30 of the present invention, including a processor 31, a memory 32, and a program or instruction stored in the memory 32 and executable on the processor 31. When the program or instruction is executed by the processor, the steps in any of the rule determination methods for realizing the computing power of the intelligent computing center of the present invention are implemented.

[0139] The present invention provides a readable storage medium with a program or instruction stored thereon. When the program or instruction is executed by a processor, each process of the embodiments of the rule determination method for realizing the computing power of the intelligent computing center as described above is implemented, and the same technical effects can be achieved. To avoid repetition, it will not be elaborated here.

[0140] Among them, the readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc. In some examples, the readable storage medium may be a non-transitory readable storage medium.

[0141] The present invention also provides a computer program product including computer instructions. When the computer instructions are executed by a processor, each process of the embodiments of the rule determination method for realizing the computing power of the intelligent computing center as described above is implemented, and the same technical effects can be achieved. To avoid repetition, it will not be elaborated here.

[0142] 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 spirit and scope of the present invention as protected by the claims, and all of them fall within the protection scope of the present invention.

Claims

1. A method for determining rules for realizing computing power of an intelligent computing center, characterized in that: include: Step S1: receiving a task to be calculated; Step S2: In response to the task to be calculated, extract the problem features that characterize the reasoning task characteristics from the task to be calculated; and obtain the rule information of each candidate rule in the preset candidate rule set, call the feature extraction large language model deployed in the intelligent computing center to infer the rule information, and obtain the rule features of each candidate rule; Step S3: vectorizing the problem feature to obtain a problem feature vector; vectorizing the rule feature to obtain a rule feature vector of each of the candidate rules; Step S4: determining a target rule for executing the task to be calculated from the set of candidate rules according to the cosine similarity between the problem feature vector and each of the rule feature vectors.

2. The method for determining rules for realizing computing power of an intelligent computing center according to claim 1, characterized in that: The feature extraction large language model includes: a code-specific feature extraction large language model and a general large language model; The feature extraction large language model deployed in the intelligent computing center is called to infer the rule information to obtain the rule features of each of the candidate rules, including: Step S21: judging the data type of the rule information to obtain a judgment result; Step S22: if the judgment result indicates that the rule information includes information in the form of rule codes, the information in the form of rule codes is input into the code-specific feature extraction large language model to obtain a first rule feature of the information in the form of rule codes; Step S23: if the judgment result indicates that the rule information includes information in the form of text description, generate a prompt word for instructing the large language model to obtain features from the information in the form of text description, input the prompt word and the information in the form of text description into the universal large language model, and obtain a second rule feature of the information in the form of text description; Step S24: combining the first rule feature and the second rule feature to obtain the rule feature.

3. The method for determining rules for realizing computing power of an intelligent computing center according to claim 2, characterized in that: The code-specific feature extraction large language model includes at least one of the following models: CodeBERT model, UniXCoder model; and / or, The universal large language model includes at least one of the following models: DeepSeek model and ChatGLM model.

4. The method for determining rules for realizing computing power of an intelligent computing center according to claim 1, characterized in that: The rule information includes at least one of the following: Programming language code, abstract syntax tree AST of code, macro-statistical characteristics of code, pseudocode, papers, manuals, hyperparameters of rules, execution process of rules, model structure of rules, loss function of rules.

5. The method for determining rules for realizing computing power of an intelligent computing center according to claim 1, characterized in that: The problem characteristics include at least one of the following: The scale of input data required to execute the task to be calculated, the type of input data required to execute the task to be calculated, and the properties and constraints of the problem to be solved by the task to be calculated.

6. The method for determining rules for realizing computing power of an intelligent computing center according to claim 1, characterized in that: The step S4 comprises: Step S41: for each of the rule feature vectors, the problem feature vector, the rule feature vector and the corresponding cosine similarity are concatenated to obtain a candidate combination; Step S42: inputting all the candidate combinations into the multi-layer perceptron MLP model deployed in the intelligent computing center, obtaining the matching degree between the task to be calculated and each of the candidate rules, and obtaining the predicted execution effect of the task to be calculated after being executed according to each of the candidate rules; Step S43: Determine the target rule according to the matching degree and the predicted execution effect.

7. A rule determination device implemented by intelligent computing center computing power, characterized in that: include: A receiving module, used for receiving tasks to be calculated; An execution module is used to extract, in response to the task to be calculated, a problem feature representing the characteristics of the reasoning task from the task to be calculated; and obtain rule information of each candidate rule in a preset candidate rule set, call a feature extraction large language model deployed in an intelligent computing center to infer the rule information, and obtain rule features of each candidate rule; A vectorization module, used for vectorizing the problem feature to obtain a problem feature vector; Vectorizing the rule features to obtain the rule feature vectors of each of the candidate rules; A determination module is used to determine a target rule for executing the task to be calculated from the set of candidate rules according to the cosine similarity between the problem feature vector and each of the rule feature vectors.

8. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps in the rule determination method for realizing the computing power of the intelligent computing center as described in any one of claims 1 to 6 are implemented.

9. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the steps in the rule determination method for realizing the computing power of the intelligent computing center as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of the rule determination method for realizing the computing power of the intelligent computing center as described in any one of claims 1 to 6.

Images