A large language model reasoning method and device based on the computing power of intelligent computing center

Abstract

The present application relates to the technical field of computing power infrastructure, and provides a large language model reasoning method and device based on the computing power of an intelligent computing center, including: determining a corresponding checklist based on a task to be reasoned; inputting the task to be reasoned into the large language model to obtain the current reasoning result; step S3: inputting the checklist, the current reasoning result, and the task to be reasoned into the large language model to obtain the current inspection result, step S4: if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, then inputting the task to be reasoned and the current inspection result into the large language model to regenerate the current reasoning result; repeating steps S3 and S4 until each sub-inspection result in the current inspection result matches the current reasoning result and the inspection item; using the current reasoning result that matches each inspection item as the target output of the large language model. In this way, the accuracy and effectiveness of large language model reasoning can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of computing power infrastructure, and in particular, to a large language model reasoning method and device based on the computing power of an intelligent computing center. Background Art

[0002] With the development of artificial intelligence technology and computing power technology, the concepts of intelligent computing centers and large language models have emerged. Intelligent computing centers refer to facilities that provide the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning) by using large-scale heterogeneous computing power resources, including general computing power and intelligent computing power. Intelligent computing centers cover facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0003] A large language model refers to a large language model (LLM), which is a language model with a large parameter scale. It is designed to understand and generate human language. It is trained with a large amount of text data and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0004] In practical applications, the reasoning process based on large language models often faces some challenges, especially when dealing with complex tasks or with multiple constraints. When generating output, large language models often "miss this and miss that", ignoring certain key factors or constraints, resulting in output results that do not meet expectations. For example, when the prompt contains multiple detailed requirements, the large language model cannot fully understand and integrate all the information, thus affecting the quality of task completion.

[0005] Especially in scenarios involving specific workflows or rules, the reasoning ability of large language models will be limited. Take the scenario where an evaluation engineer is involved as an example. Suppose there is a clear rule that requires an email notification to be sent when an evaluation engineer is involved. This requirement is listed as an important constraint in the prompt, but due to the uneven distribution of attention of the large language model during reasoning, it may fail to pay enough attention to this key rule, thereby missing the necessary step of sending an email. This situation not only affects the completion of the task, but may also cause confusion in the subsequent workflow, thereby affecting the progress of the overall project.

[0006] In summary, although large language models have significant advantages in the field of natural language processing, they ignore certain key factors or constraints during the reasoning process, resulting in output results that do not meet expectations, reducing their effectiveness and accuracy in processing tasks. Summary of the invention

[0007] The embodiments of the present application provide a large language model reasoning method and device based on the computing power of an intelligent computing center to solve the technical problem that the current large language model ignores certain key factors or constraints during the reasoning process, resulting in the output results not meeting expectations, thereby reducing its effectiveness and accuracy in processing tasks.

[0008] In order to solve the above technical problems, this application is implemented as follows:

[0009] In a first aspect, an embodiment of the present application provides a large language model reasoning method based on the computing power of an intelligent computing center, the method comprising:

[0010] Step S1: determining a corresponding checklist based on the task to be inferred; wherein the checklist includes a plurality of check items;

[0011] Step S2: input the task to be inferred into the large language model to obtain the current inference result;

[0012] Step S3: inputting the inspection list, the current reasoning result and the task to be reasoned into the large language model to obtain the current inspection result, wherein the current inspection result is used to record the sub-inspection result of inspecting the current reasoning result based on each inspection item;

[0013] Step S4: if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, the task to be reasoned and the current inspection result are input into the large language model to regenerate the current reasoning result;

[0014] Step S5: Repeat steps S3 and S4 until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item;

[0015] Step S6: taking the current inference result that matches each check item as the target output of the large language model.

[0016] Optionally, step S1 includes:

[0017] Step S11: input the task to be inferred into the large language model, and output the checklist corresponding to the task to be inferred.

[0018] Optionally, before step S11, the method further includes:

[0019] Step S1a: Obtain samples of tasks to be reasoned and samples of checklists;

[0020] Step S1b: The sample of the task to be inferred and the sample of the checklist are used as training data to train the large language model. The trained large language model can output a corresponding checklist based on the task to be inferred.

[0021] Optionally, step S1 includes:

[0022] Step S12: Determine relevant information of the task to be inferred;

[0023] Step S13: generating a check item corresponding to the task to be inferred based on the relevant information and preset rules;

[0024] Step S14: Merge the inspection items to obtain the inspection list, wherein the preset rules include the matching relationship between the relevant information and the inspection items, and the relevant information includes at least one of the following: the task type of the task to be reasoned, the constraint conditions of the task to be reasoned, and the target requirements of the task to be reasoned.

[0025] In a second aspect, an embodiment of the present application provides a large language model reasoning device based on the computing power of an intelligent computing center, the device comprising:

[0026] A determination module, configured to determine a corresponding checklist based on the task to be inferred; wherein the checklist includes a plurality of check items;

[0027] An execution module, used for inputting the task to be inferred into the large language model to obtain a current inference result;

[0028] Inputting the inspection list, the current reasoning result and the task to be reasoned into the large language model to obtain a current inspection result, wherein the current inspection result is used to record a sub-inspection result of inspecting the current reasoning result based on each inspection item;

[0029] If there is at least one sub-check result that the current reasoning result does not match the check item, input the task to be reasoned and the current check result into the large language model to regenerate the current reasoning result;

[0030] Repeat the step of inputting the inspection list, the current reasoning result, and the task to be reasoned into the large language model to obtain the current inspection result; if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, inputting the task to be reasoned and the current inspection result into the large language model to regenerate the current reasoning result, until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item;

[0031] The current inference result that matches each check item is used as the target output of the large language model.

[0032] Optionally, the determination module is further used to input the task to be inferred into the large language model, and output the checklist corresponding to the task to be inferred.

[0033] Optionally, the execution module is further used to obtain a sample of the task to be inferred and a sample of the checklist before inputting the task to be inferred into the large language model and outputting the checklist corresponding to the task to be inferred;

[0034] The large language model is trained by using the task samples to be inferred and the checklist samples as training data. The trained large language model can output a corresponding checklist based on the task to be inferred.

[0035] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of a large language model reasoning method based on the computing power of an intelligent computing center as described in the first aspect are implemented.

[0036] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a large language model reasoning method based on the computing power of an intelligent computing center as described in the first aspect are implemented.

[0037] In a fifth aspect, an embodiment of the present application provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of a large language model reasoning method based on the computing power of an intelligent computing center as described in the first aspect.

[0038] Therefore, by utilizing the large-scale heterogeneous computing resources provided by the Intelligent Computing Center, strong computing power support can be provided for the reasoning process of the large language model. The large language model can perform a large number of reasoning and checks in a shorter time, thereby speeding up the model iteration. The high-performance computing capabilities of the Intelligent Computing Center enable the large language model to handle multiple tasks and check items at the same time, and quickly evaluate different reasoning strategies. This parallel computing capability not only improves efficiency, but also enhances the flexibility and adaptability of the large language model in complex tasks. In addition, the checklist is used to assist the large language model in reasoning, so that the large language model can have a clear guidance framework during the reasoning process, ensuring that important information is not missed when generating outputs, which can significantly reduce the error rate of the large language model during reasoning and improve the accuracy and effectiveness of the large language model reasoning.

[0039] In summary, combining the checklist idea with the computing power resources of the intelligent computing center to assist the reasoning process of the large language model can not only improve the reasoning speed and processing power of the large language model, but also provide a clear guidance structure in the reasoning process of the large language model, so that it can systematically consider all relevant factors and constraints when reasoning, thereby effectively solving the omission problem in the reasoning process of the large language model and improving the effectiveness and accuracy of the reasoning of the large language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0041] Figure 1 A flowchart of a large language model reasoning method based on the computing power of an intelligent computing center provided in an embodiment of the present application;

[0042] Figure 2 A structural block diagram of a large language model inference device based on the computing power of an intelligent computing center provided in an embodiment of the present application;

[0043] Figure 3 A structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] The following is a brief description of the technical terms involved in the present invention.

[0046] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0047] The "computing power" (CP) described in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, super computing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations performed per second (FLOPS: Floating Point Operations Per Second, 1EFLOPS = 10^18FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1EFLOPS is about the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP = CP general + CP intelligent + CP super.

[0048] The "Network Power" (NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, etc. The carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the memory bandwidth.

[0049] The "Storage Power" (SP) described in the present invention is the comprehensive ability of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB = 2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability.

[0050] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, security and reliability, and green and low-carbon.

[0051] The “computing power” mentioned in the present invention includes general computing power, intelligent computing power and super computing power.

[0052] The “general computing power” mentioned 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.

[0053] The "intelligent computing power" described in the present invention is a computing platform for large-scale deployment of special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative applications of artificial intelligence, such as natural language processing and machine vision.

[0054] The "super computing power" mentioned in the present invention is 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 uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0055] The "intelligent computing center" described in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning scenarios) by using large-scale heterogeneous computing power resources, including general computing power (CPU: Central Processing Unit) and intelligent computing power (GPU: Graphics Processing Unit, FPGA: Field Programmable Gate Array, ASIC: Application Specific Integrated Circuit, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0056] The "computing resources" mentioned in the present invention refer to the technologies and facilities with information calculation, transmission, storage and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), 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 supporting and guarantee resources such as wind, fire, water and electricity.

[0057] The “large language model” mentioned in the present invention refers to a large language model (LLM), which is a language model with a large parameter scale, designed to understand and generate human language, trained with a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0058] The "checklist" of the present invention is a tool for helping to organize and check tasks, matters or steps, usually listing a series of check items to be completed in the form of a list. Its purpose is to ensure that there are no omissions or errors when carrying out a certain activity, task or process, and to complete the task or project accurately.

[0059] Figure 1 A large language model reasoning method based on the computing power of an intelligent computing center provided according to an embodiment of the present application is shown. Figure 1 As shown, the method includes:

[0060] Step S1: Determine a corresponding checklist based on the task to be reasoned;

[0061] Among them, the checklist includes multiple check items;

[0062] Step S2: Input the task to be inferred into the large language model to obtain the current inference result;

[0063] Step S3: input the inspection list, the current reasoning result, and the task to be reasoned into the large language model to obtain the current inspection result;

[0064] The current inspection result is used to record the sub-inspection result of inspecting the current reasoning result based on each inspection item;

[0065] Step S4: if there is at least one sub-check result that the current reasoning result does not match the check item, the task to be reasoned and the current check result are input into the large language model to regenerate the current reasoning result;

[0066] Step S5: Repeat steps S3 and S4 until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item;

[0067] Step S6: The current inference result that matches each check item is used as the target output of the large language model.

[0068] In step S1, it is necessary to clarify the task to be reasoned and its related requirements, create a checklist based on the nature and objectives of the task to be reasoned, and list all necessary check items. The check items may include specific requirements, standards, steps, precautions, etc. of the task.

[0069] In a possible implementation, step S1 includes: step S11: inputting the task to be inferred into the large language model, and outputting a checklist corresponding to the task to be inferred. Before step S11, the method further includes: step S1a: obtaining samples of the task to be inferred and samples of the checklist; step S1b: using the samples of the task to be inferred and the checklist as training data to train the large language model, and the trained large language model can output a corresponding checklist based on the task to be inferred.

[0070] That is to say, a checklist corresponding to the task to be inferred can be generated through a large language model. First, sample data related to the task to be inferred can be collected, and the collected samples of the task to be inferred and the corresponding checklist samples can be used as training data to train the model. The goal of the training is to enable the large language model to understand and generate a checklist that matches a specific task to be inferred. After training, the model will have the ability to automatically generate a corresponding checklist based on the input task to be inferred. This ability enables the model to quickly and accurately provide users with the required checklists in practical applications, improving work efficiency. After completing the training, users can input new tasks to be inferred into the trained large language model, and the model will generate a checklist related to the task based on the knowledge it has learned. For example: In the software development process, developers need to ensure code quality and functional integrity. Developers can enter the description of new functions, and the large language model will generate corresponding test checklists, including unit testing, integration testing, user acceptance testing, and code review, which ensures that all key functions are fully tested before release.

[0071] As a result, the automatic generation of checklists from the task to be inferred is realized, allowing users to quickly obtain a checklist for a specific task. This not only reduces the time and effort of manually creating checklists, but also ensures the systematic and consistent generation results.

[0072] In one possible implementation, step S1 includes: step S12: determining relevant information of the task to be reasoned; step S13: generating inspection items corresponding to the task to be reasoned based on the relevant information and preset rules; step S14: merging the inspection items to obtain an inspection list, wherein the preset rules include a matching relationship between the relevant information and the inspection items, and the relevant information includes at least one of the following: the task type of the task to be reasoned, the constraints of the task to be reasoned, and the target requirements of the task to be reasoned.

[0073] It should be noted that in this step, it is necessary to collect and identify relevant information of the task to be reasoned. This information may include: task type: such as project management, software development, etc.; constraints: such as time constraints, resource constraints, etc.; target requirements: such as expected results, quality standards, customer needs, etc. This information provides necessary information for the subsequent generation of inspection items; after that, specific inspection items can be generated according to the relevant information determined in step S12, combined with preset rules (such as the matching relationship between relevant information and inspection items). The preset rules may include: standard inspection items corresponding to different task types, the impact of constraints on inspection items (for example, time constraints may require a more compact inspection process), and target requirements determine the priority and importance of inspection items; finally, the inspection items can be merged to obtain a checklist, which will cover all key inspection items related to the task to be reasoned to ensure the comprehensiveness and effectiveness of task execution.

[0074] It should be noted that of the two implementation methods mentioned above, the first method relies on a well-trained large language model to directly generate a checklist based on the input task to be inferred. This process mainly depends on the generation capability of the model and the quality of the training data. The second implementation method uses systematic steps to first determine the relevant information, then generate check items based on preset rules, and finally merge them into a checklist. This process relies more on clear logic and rules, rather than relying entirely on the generation capability of the model. You can choose the appropriate checklist generation method based on the specific application scenario.

[0075] In step S2, the task to be inferred needs to be input into the large language model to obtain the current inference result. This step utilizes the generation capability of the large language model and combines the task requirements to produce task-related outputs. This is the starting point of the entire process and lays the foundation for subsequent inspection and verification.

[0076] In step S3, the current reasoning result generated in step S2, the checklist generated in step S1, and the original task to be reasoned are input into the large language model together to check the current reasoning result. The large language model will evaluate the current reasoning result one by one according to the check items in the checklist and in combination with the task to be reasoned, and generate the corresponding current inspection result. The current inspection result is used to record the matching of each inspection item with the current reasoning result, that is, whether the current reasoning result passes the inspection of each inspection item in the checklist. Therefore, by comparing the current reasoning result with the checklist based on the large language model, it is possible to systematically identify which parts of the current reasoning result meet the requirements and which parts do not meet the requirements. It should be noted that in the comparison and inspection, the current reasoning result can be verified item by item according to each item in the checklist, and it can also be checked in parallel, that is, multiple outputs or multiple inspection items are verified at the same time to improve the verification efficiency.

[0077] In step S4, if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, the task to be reasoned and the current inspection result are input into the large language model to regenerate the current reasoning result. In step S5, steps S3 and S4 need to be repeated until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item.

[0078] In other words, the large language model can use the inspection results as feedback to adjust its reasoning strategy and ensure that all relevant constraints and conditions are taken into account as much as possible when generating new reasoning results. This not only helps to correct the current errors, but also enhances the accuracy and reliability of the large language model in subsequent reasoning. As a result, the large language model can optimize its generation capabilities based on feedback information, gradually improve the quality and accuracy of generating new reasoning results, until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item. This iterative process helps the large language model better adapt to the task requirements until the current reasoning result output by the large language model can pass the inspection of all inspection items.

[0079] In step S6, the current reasoning result that matches each check item is used as the target output of the large language model, thereby ensuring that the output generated by the large language model not only meets the requirements of the task to be reasoned, but also satisfies all the check items in the checklist. This method improves the effectiveness and accuracy of the large language model output.

[0080] In summary, by introducing the checklist into the reasoning of the large language model, the reasoning quality of the large language model in complex tasks can be effectively improved, errors and omissions of key information can be reduced, the comprehensiveness, accuracy and effectiveness of the output of the large language model can be ensured, and the adaptability and reliability of the large language model can be enhanced. In addition, the checklist can be generated specifically according to the different tasks to be reasoned, and can be applied to the reasoning of various specific tasks, expanding the application scenarios.

[0081] Moreover, by utilizing the large-scale heterogeneous computing resources provided by the Intelligent Computing Center, strong computing power support can be provided for the reasoning process of the large language model. The large language model can perform a large number of reasoning and checks in a shorter time, thereby speeding up the model iteration. The high-performance computing capabilities of the Intelligent Computing Center enable the large language model to handle multiple tasks and check items at the same time, and quickly evaluate different reasoning strategies. This parallel computing capability not only improves efficiency, but also enhances the flexibility and adaptability of the large language model in complex tasks. In addition, the reasoning of the large language model is assisted by the checklist, so that the large language model can have a clear guidance framework during the reasoning process, ensuring that important information is not missed when generating outputs, which can significantly reduce the error rate of the large language model during reasoning and improve the accuracy and effectiveness of the large language model reasoning.

[0082] Figure 2 A large language model reasoning device based on the computing power of an intelligent computing center provided according to an embodiment of the present application is shown. Figure 2 As shown, the device comprises:

[0083] A determination module 201 is used to determine a corresponding checklist based on the task to be inferred; wherein the checklist includes a plurality of check items;

[0084] An execution module 202 is used to input the task to be inferred into the large language model to obtain the current inference result;

[0085] Input the inspection list, the current reasoning result, and the task to be reasoned into the large language model to obtain the current inspection result, wherein the current inspection result is used to record the sub-inspection result of the inspection of the current reasoning result based on each inspection item;

[0086] If there is at least one sub-check result that the current reasoning result does not match the check item, the task to be reasoned and the current check result are input into the large language model to regenerate the current reasoning result;

[0087] Repeat the steps of inputting the checklist, the current reasoning result, and the task to be reasoned into the large language model to obtain the current check result; if there is at least one sub-check result for which the current reasoning result does not match the check item, inputting the task to be reasoned and the current check result into the large language model to regenerate the current reasoning result, until each sub-check result in the current check result matches the current reasoning result and the check item;

[0088] The current inference result that matches each check item is used as the target output of the large language model.

[0089] In one possible implementation,

[0090] The determination module 201 is further used to input the task to be inferred into the large language model and output a checklist corresponding to the task to be inferred.

[0091] In one possible implementation,

[0092] The execution module 202 is further used to obtain a sample of the task to be inferred and a sample of the checklist before inputting the task to be inferred into the large language model and outputting a checklist corresponding to the task to be inferred;

[0093] The samples of the task to be inferred and the checklist samples are used as training data to train the large language model. The trained large language model can output the corresponding checklist based on the task to be inferred.

[0094] In one possible implementation, the determination module 201 is also used to determine relevant information of the task to be reasoned; based on the relevant information and preset rules, generate inspection items corresponding to the task to be reasoned; merge the inspection items to obtain an inspection list, wherein the preset rules include a matching relationship between the relevant information and the inspection items, and the relevant information includes at least one of the following: the task type of the task to be reasoned, the constraint conditions of the task to be reasoned, and the target requirements of the task to be reasoned.

[0095] Therefore, by utilizing the large-scale heterogeneous computing resources provided by the Intelligent Computing Center, strong computing power support can be provided for the reasoning process of the large language model. The large language model can perform a large number of reasoning and checks in a shorter time, thereby speeding up the model iteration. The high-performance computing capabilities of the Intelligent Computing Center enable the large language model to handle multiple tasks and check items at the same time, and quickly evaluate different reasoning strategies. This parallel computing capability not only improves efficiency, but also enhances the flexibility and adaptability of the large language model in complex tasks. In addition, the checklist is used to assist the large language model in reasoning, so that the large language model can have a clear guidance framework during the reasoning process, ensuring that important information is not missed when generating outputs, which can significantly reduce the error rate of the large language model during reasoning and improve the accuracy and effectiveness of the large language model reasoning.

[0096] In summary, combining the checklist idea with the computing power resources of the intelligent computing center to assist the reasoning process of the large language model can not only improve the reasoning speed and processing power of the large language model, but also provide a clear guidance structure in the reasoning process of the large language model, so that it can systematically consider all relevant factors and constraints when reasoning, thereby effectively solving the omission problem in the reasoning process of the large language model and improving the effectiveness and accuracy of the reasoning of the large language model.

[0097] The present application embodiment provides an electronic device 30, such as Figure 3 As shown, the electronic device 30 includes: a processor 301, a memory 302, and a program stored in the memory 302 and executable on the processor 301. When the program is executed by the processor 301, the steps of a large language model reasoning method based on the computing power of an intelligent computing center as shown in the above embodiment are implemented.

[0098] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a large language model reasoning method based on the computing power of an intelligent computing center as shown in the above embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0099] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the above-mentioned large language model reasoning method based on the computing power of an intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0100] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0101] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0102] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A large language model reasoning method based on the computing power of an intelligent computing center, characterized in that: The method comprises: Step S1: determining a corresponding checklist based on the task to be inferred; wherein the checklist includes a plurality of check items; Step S2: input the task to be inferred into the large language model to obtain the current inference result; Step S3: inputting the inspection list, the current reasoning result and the task to be reasoned into the large language model to obtain the current inspection result, wherein the current inspection result is used to record the sub-inspection result of inspecting the current reasoning result based on each inspection item; Step S4: if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, the task to be reasoned and the current inspection result are input into the large language model to regenerate the current reasoning result; Step S5: Repeat steps S3 and S4 until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item; Step S6: taking the current inference result that matches each check item as the target output of the large language model.

2. The method according to claim 1, characterized in that Step S1 includes: Step S11: input the task to be inferred into the large language model, and output the checklist corresponding to the task to be inferred.

3. The method according to claim 2, characterized in that Before step S11, the method further includes: Step S1a: Obtain samples of tasks to be reasoned and samples of checklists; Step S1b: The large language model is trained by using the task samples to be inferred and the checklist samples as training data. The trained large language model can output a corresponding checklist based on the task to be inferred.

4. The method according to claim 1, characterized in that: Step S1 includes: Step S12: Determine relevant information of the task to be inferred; Step S13: generating a check item corresponding to the task to be inferred based on the relevant information and preset rules; Step S14: Merge the inspection items to obtain the inspection list, wherein the preset rules include the matching relationship between the relevant information and the inspection items, and the relevant information includes at least one of the following: the task type of the task to be reasoned, the constraint conditions of the task to be reasoned, and the target requirements of the task to be reasoned.

5. A large language model reasoning device based on the computing power of an intelligent computing center, characterized in that: The device comprises: A determination module, configured to determine a corresponding checklist based on the task to be inferred; wherein the checklist includes a plurality of check items; An execution module, used for inputting the task to be inferred into the large language model to obtain a current inference result; Inputting the inspection list, the current reasoning result and the task to be reasoned into the large language model to obtain a current inspection result, wherein the current inspection result is used to record a sub-inspection result of inspecting the current reasoning result based on each inspection item; If there is at least one sub-inspection result that the current reasoning result does not match the inspection item, input the task to be reasoned and the current inspection result into the large language model to regenerate the current reasoning result; Repeat the step of inputting the inspection list, the current reasoning result, and the task to be reasoned into the large language model to obtain the current inspection result; if there is at least one sub-inspection result that the current reasoning result does not match the inspection item, inputting the task to be reasoned and the current inspection result into the large language model to regenerate the current reasoning result, until each sub-inspection result in the current inspection result matches the current reasoning result with the inspection item; The current inference result that matches each check item is used as the target output of the large language model.

6. The device according to claim 5, characterized in that The determination module is further used to input the task to be inferred into the large language model and output the checklist corresponding to the task to be inferred.

7. The device according to claim 6, characterized in that The execution module is further configured to obtain a sample of the task to be inferred and a sample of the checklist before inputting the task to be inferred into the large language model and outputting the checklist corresponding to the task to be inferred; The large language model is trained by using the task samples to be inferred and the checklist samples as training data. The trained large language model can output a corresponding checklist based on the task to be inferred.

8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of a large language model reasoning method based on the computing power of an intelligent computing center are implemented as described in any one of claims 1 to 4.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a large language model reasoning method based on the computing power of an intelligent computing center as described in any one of claims 1 to 4.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of a large language model reasoning method based on the computing power of an intelligent computing center as described in any one of claims 1-4.

Images