High-performance computing model optimization system based on artificial intelligence
Through the collaborative work of multi-module optimization systems of artificial intelligence optimization systems, the problems of insufficient comprehensive optimization of traditional systems and unclear indicators have been solved, and the multi-dimensional evaluation and optimization solutions for the optimization of large computing power models have been achieved.
Patent Information
- Application Number
- CN202411368581.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-29
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-29
AI Technical Summary
The traditional large computing power model optimization system lacks standardized judgment of optimization indicators, and the optimization is not comprehensive enough. It only performs limitation optimization of model training and inference processes, and cannot clearly understand the optimization effect.
Design a large computing power model optimization system based on artificial intelligence, including a number of runs determination module, an optimization data acquisition module, an optimization data preprocessing module, an optimization data analysis module and an integrated data analysis module. Through real-time monitoring and data analysis, test reports are generated. Determine whether the optimization capabilities meet the standards and provide improvement solutions.
A comprehensive evaluation of the large computing power model optimization system is achieved, multi-dimensional evaluation of computing cost, model stability, training process and inference process is provided, testing reports are generated, and optimization solutions are supported.
Smart Images

Figure CN118885785B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to a large computing power model optimization system based on artificial intelligence. Background Art
[0002] The traditional large computing power model optimization system includes computing power infrastructure optimization module, data management and optimization module, model training optimization module, model reasoning optimization module, system monitoring and management module, and automation and intelligent operation and maintenance module. The computing power infrastructure optimization module is used to select and optimize hardware and build a distributed computing framework; the data management and optimization module is used to collect and preprocess data, adopt efficient storage solutions, and optimize data transmission; the model training optimization module uses parallel training technology and optimizes memory and storage; the model reasoning optimization module is used to compress and prune models; the system monitoring and management module uses cluster management tools and real-time monitoring of the operating status and performance indicators of the computing power center; and the automation and intelligent operation and maintenance module is used to perform automated deployment and configuration and intelligent operation and maintenance. The traditional large computing power model optimization system effectively improves the efficiency of model training and reasoning, accelerates technological innovation and industrial upgrading, promotes sustainable development, and promotes the development of the open source ecosystem.
[0003] However, there are still some shortcomings in its actual use. For example, the traditional large-computing power model optimization system only optimizes the large-computing power model, and does not provide standards for judging the indicators of the large-computing power model optimization system. It is impossible to clearly and intuitively understand whether the optimization of the large-computing power model meets the requirements; the optimization data of the traditional large-computing power model optimization system is not comprehensive enough, and it only optimizes the model training process and the model reasoning process. The optimized nodes and parameters are too limited and not comprehensive enough. Summary of the invention
[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a large computing power model optimization system based on artificial intelligence, which solves the problems raised in the above-mentioned background technology through the following scheme.
[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: A large computing power model optimization system based on artificial intelligence, including a system operation database, a system central processing module and a user information terminal, and also including an operation number determination module, an optimization data acquisition module, an optimization data preprocessing module, an optimization data analysis module, a comprehensive data analysis module and a human-computer interaction module;
[0006] The system operation database includes all data texts of the large computing power model optimization system, and collects the information texts output by each module in real time. The system central processing module is used to centrally control the information text instructions output by each module, and the user information terminal is an information output device for receiving the large computing power model optimization system;
[0007] The operation number determination module is used to filter out characteristic text data from the data text, and the number of times of filtering is recorded as 1, 2...n in sequence;
[0008] The optimized data collection module selects the first key text based on the characteristic text data;
[0009] The optimized data preprocessing module is used to perform preprocessing analysis on the key text to obtain a second key text;
[0010] The optimization data analysis module obtains any text output value in the operation number determination module based on the second key text analysis;
[0011] The comprehensive data analysis module obtains a comprehensive text output value based on any text output value in the optimized data analysis module;
[0012] The human-computer interaction module outputs information based on the comprehensive text output value and the user information terminal, determines the optimization capability of the large computing power model optimization system, and generates a test report.
[0013] Preferably, the method for filtering out the characteristic text data is to monitor the operating status of the large computing power model system in real time, and use a log tool to record the data during each system operation, and collect the characteristic text data during the operation of the large computing power model optimization system through characteristic word screening.
[0014] Preferably, the user information terminal is bound to a mobile phone or any electronic device capable of output based on the administrator's work information.
[0015] Preferably, the first key text includes calculation cost parameters, model stability parameters, training process optimization parameters and reasoning process optimization parameters, and the calculation cost parameters include the number of model layers, model dimensions, number of shared parameters, number of quantization parameters and data compression ratio, which are marked as L M , D M 、N S 、N Q and R D ; Model stability parameters include signal-to-noise ratio, model outliers, number of training data, and model complexity, which are marked as R S 、M O 、N T and C M ; The optimization parameters of the training process include the degree of reduction in training accuracy, the number of parallel nodes added, and the number of gradient checkpoints, which are marked as R A 、N P and N C ; The optimization parameters of the inference process include the number of key-value pairs stored, the degree of hardware acceleration, and the number of pruned neurons, which are marked as N K , DH and N E .
[0016] Preferably, the second key text is obtained by constructing an optimized data preprocessing model, importing the collected computing cost parameters, model stability parameters, training process optimization parameters and reasoning process optimization parameters into the optimized data preprocessing model, and calculating the second key text, including a computing cost evaluation value, a model stability evaluation value, a training process optimization evaluation value and a reasoning process optimization evaluation value, wherein the computing cost evaluation value is specifically expressed as:
[0017] ,
[0018] A C represents the computational cost evaluation value of the target large computing power model optimization system, L M represents the number of model layers of the target large computing power model optimization system, D M N represents the model dimension of the target large computing power model optimization system. S Represents the number of shared parameters of the target large computing model optimization system, N Q Represents the number of quantitative parameters of the target large computing model optimization system, R Di represents the data compression ratio of the target large computing power model optimization system during the i-th operation, δ1 and δ2 represent the influence coefficients of the number of model layers and model dimensions on the computational cost evaluation value, respectively, and b1 and b2 are known constants; the model stability evaluation value is specifically expressed as:
[0019] ,
[0020] A S represents the model stability evaluation value of the target large computing power model optimization system, R Si represents the signal-to-noise ratio of the target large computing power model optimization system during the i-th operation, M O Represents the model outlier value of the target large computing power model optimization system, N Ti represents the amount of training data during the i-th operation of the target large computing power model optimization system, C M represents the model complexity of the target large computing power model optimization system, δ3 represents the influence coefficient of model complexity on the model stability evaluation value, and b3 is a known constant; the optimization evaluation value of the training process is specifically expressed as:
[0021] ,
[0022] A T represents the optimization evaluation value of the training process of the target large computing power model optimization system, R Ai N represents the degree of reduction in training accuracy during the i-th operation of the target large computing power model optimization system. PiN represents the number of parallel nodes added during the i-th operation of the target large computing power model optimization system. Ci represents the number of gradient checkpoints during the i-th operation of the target large computing power model optimization system, and C1 represents a known constant; the optimization evaluation value of the inference process is specifically expressed as:
[0023] ,
[0024] A R represents the optimization evaluation value of the inference process of the target large computing power model optimization system, N Ki represents the number of key-value pairs stored during the i-th operation of the target large computing power model optimization system, D H Indicates the hardware acceleration level of the target large computing model optimization system, N Ei It represents the number of pruned neurons during the i-th operation of the target large computing power model optimization system, and C2 and C3 represent known constants.
[0025] Preferably, the method for calculating the arbitrary text output value is to establish an optimized data analysis model, and import the second key text into the optimized data analysis model to calculate the arbitrary text output value, including the model structure optimization evaluation value and the model process optimization evaluation value, and the model structure optimization evaluation value is specifically expressed as: ,φ1 represents the model structure optimization evaluation value of the target large computing power model optimization system; the model process optimization evaluation value is specifically expressed as: , φ2 represents the model process optimization evaluation value of the target large computing power model optimization system.
[0026] Preferably, the method for calculating the comprehensive text output value is to establish a comprehensive data analysis model, import any text output value into the comprehensive data analysis model, and calculate the comprehensive text output value, which is specifically expressed as:
[0027] ,
[0028] φ represents the comprehensive text output value of the target large computing power model optimization system, φ1' represents the standard value of the model structure optimization evaluation value of the target large computing power model optimization system, and φ2' represents the standard value of the model process optimization evaluation value of the target large computing power model optimization system.
[0029] Preferably, the method for judging the optimization capability of the large computing power model optimization system is to establish a standard value for the comprehensive text output value and mark it as φ 标 , when φ<φ 标 When φ≥φ 标, it indicates that the comprehensive text output value is greater than or equal to the standard value of the comprehensive text output value, and the optimization capability of the target large computing power model optimization system meets the standard.
[0030] Technical effects and advantages of the present invention:
[0031] 1. The present invention calculates the comprehensive text output value by optimizing the data preprocessing module, the optimization data analysis module and the comprehensive data analysis module, and feeds back the optimization capability of the target large computing power model optimization system through the comprehensive text output value, and compares the comprehensive text output value with the standard value of the comprehensive text output value through the human-computer interaction module to determine whether the optimization of the large computing power model by the target large computing power model optimization system meets the expected requirements. If the optimization effect does not meet the requirements, the judgment result is fed back to the target large computing power model optimization system to enable it to continue to improve the optimization plan;
[0032] 2. The present invention collects sufficient running number samples and feedbacks relevant parameters of the target large computing power model optimization system performance through the running number determination module and the optimization data collection module, and judges the optimization ability of the target large computing power model optimization system for the large computing power model from four dimensions: computing cost, model stability, training process optimization and reasoning process optimization. The collected and calculated parameters are passed to the database, and a test report is generated to provide data support for the subsequent improvement of the large computing power model optimization system. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is a schematic diagram of the overall structure of the present invention.
[0034] Figure 2 It is a diagram of the method steps of the present invention.
[0035] Figure 3 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION
[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0037] As attached Figure 1 The artificial intelligence-based large-computing-power model optimization system shown includes a system operation database, a system central processing module and a user information terminal, and also includes an operation number determination module, an optimization data acquisition module, an optimization data preprocessing module, an optimization data analysis module, a comprehensive data analysis module and a human-computer interaction module.
[0038] The output end of the operation number determination module is connected to the input end of the optimization data acquisition module, the output end of the optimization data acquisition module is connected to the input end of the optimization data preprocessing module, the output end of the optimization data preprocessing module is connected to the input end of the optimization data analysis module, the input end of the optimization data analysis module is connected to the input end of the comprehensive data analysis module, and the output end of the comprehensive data analysis module is connected to the input end of the human-computer interaction module.
[0039] The system operation database includes all data texts of the large computing power model optimization system, and collects the information text output by each module in real time. The system central processing module is used to control the information text instructions output by each module, and the user information terminal is an information output device for receiving the large computing power model optimization system.
[0040] The method for filtering out characteristic text data is to monitor the operating status of the large computing power model system in real time, and use a log tool to record data during each system operation, and collect characteristic text data during the operation of the large computing power model optimization system through characteristic word screening.
[0041] The user information terminal is bound to a mobile phone or any electronic device capable of output based on the administrator's work information.
[0042] The operation times determination module is used to filter out characteristic text data from the data text, and the times of filtering are recorded as 1, 2...n in sequence.
[0043] The optimized data collection module selects the first key text based on the characteristic text data.
[0044] The first key text includes calculation cost parameters, model stability parameters, training process optimization parameters and inference process optimization parameters. The calculation cost parameters include the number of model layers, model dimensions, number of shared parameters, number of quantized parameters and data compression ratio, which are marked as L M , D M 、N S 、N Q and R D ; Model stability parameters include signal-to-noise ratio, model outliers, number of training data, and model complexity, which are marked as R S 、M O 、N T and C M ; The optimization parameters of the training process include the degree of reduction in training accuracy, the number of parallel nodes added, and the number of gradient checkpoints, which are marked as R A 、N P and N C ; The optimization parameters of the inference process include the number of key-value pairs stored, the degree of hardware acceleration, and the number of pruned neurons, which are marked as NK , D H and N E .
[0045] In this embodiment, it is necessary to specifically explain that the model visualization tool Netron is used to detect the model layers and model dimensions of the large computing power model optimization system, and the data size before compression by the target large computing power model optimization system is marked as d f , the data size after compression by the target large computing power model optimization system is marked as d a , the data compression ratio is expressed as: R D =d a / d f ; Calculate the signal-to-noise ratio by signal strength and noise strength, use the generalized ESD test method to detect model outliers, and use TensorFlow's tf.profiler to calculate the model complexity of the high-computing power model.
[0046] In this embodiment, it is necessary to specifically explain that the degree of reduction in training accuracy is calculated by first setting the best performance of the large computing power model as the benchmark accuracy, detecting and recording the accuracy of the large computing power model during the training process, and marking the benchmark accuracy as P b The accuracy after optimization by the target large computing power model optimization system is marked as P n , the degree of reduction in training accuracy R can be calculated A =[(P b -P n ) / P b ]×100%; To measure the degree of hardware acceleration, first establish the benchmark time. The time it takes for the large computing model that has not been optimized by the target large computing model optimization system to run once is determined as the benchmark time and marked as t1. Then measure the time required for the large computing model that has not been optimized by the target large computing model optimization system to run once and mark it as t2 to calculate the degree of hardware acceleration D H =[(t1-t2) / t1]×100%.
[0047] The optimized data preprocessing module is used to perform preprocessing analysis on the key text to obtain the second key text.
[0048] The second key text is obtained by constructing an optimized data preprocessing model, importing the collected computing cost parameters, model stability parameters, training process optimization parameters, and reasoning process optimization parameters into the optimized data preprocessing model, and calculating the second key text, including a computing cost evaluation value, a model stability evaluation value, a training process optimization evaluation value, and a reasoning process optimization evaluation value. The computing cost evaluation value is specifically expressed as:
[0049] ,
[0050] A Crepresents the computational cost evaluation value of the target large computing power model optimization system, L M represents the number of model layers of the target large computing power model optimization system, D M N represents the model dimension of the target large computing power model optimization system. S Represents the number of shared parameters of the target large computing model optimization system, N Q Represents the number of quantitative parameters of the target large computing model optimization system, R Di represents the data compression ratio of the target large computing power model optimization system during the i-th operation, δ1 and δ2 represent the influence coefficients of the number of model layers and model dimensions on the computational cost evaluation value, respectively, and b1 and b2 are known constants; the model stability evaluation value is specifically expressed as:
[0051] ,
[0052] A S represents the model stability evaluation value of the target large computing power model optimization system, R Si represents the signal-to-noise ratio of the target large computing power model optimization system during the i-th operation, M O Represents the model outlier value of the target large computing power model optimization system, N Ti represents the amount of training data during the i-th operation of the target large computing power model optimization system, C M represents the model complexity of the target large computing power model optimization system, δ3 represents the influence coefficient of model complexity on the model stability evaluation value, and b3 is a known constant; the optimization evaluation value of the training process is specifically expressed as:
[0053] ,
[0054] A T represents the optimization evaluation value of the training process of the target large computing power model optimization system, R Ai N represents the degree of reduction in training accuracy during the i-th operation of the target large computing power model optimization system. Pi N represents the number of parallel nodes added during the i-th operation of the target large computing power model optimization system. Ci represents the number of gradient checkpoints during the i-th operation of the target large computing power model optimization system, and C1 represents a known constant; the optimization evaluation value of the inference process is specifically expressed as:
[0055] ,
[0056] A R represents the optimization evaluation value of the inference process of the target large computing power model optimization system, N Ki represents the number of key-value pairs stored during the i-th operation of the target large computing power model optimization system, D H Indicates the hardware acceleration level of the target large computing model optimization system, NEi It represents the number of pruned neurons during the i-th operation of the target large computing power model optimization system, and C2 and C3 represent known constants.
[0057] The optimization data analysis module obtains any text output value in the running times determination module based on the second key text analysis.
[0058] The method for calculating the arbitrary text output value is to establish an optimized data analysis model, and import the second key text into the optimized data analysis model to calculate the arbitrary text output value, including the model structure optimization evaluation value and the model process optimization evaluation value. The model structure optimization evaluation value is specifically expressed as: ,φ1 represents the model structure optimization evaluation value of the target large computing power model optimization system; the model process optimization evaluation value is specifically expressed as: , φ2 represents the model process optimization evaluation value of the target large computing power model optimization system.
[0059] The comprehensive data analysis module obtains a comprehensive text output value based on any text output value in the optimized data analysis module.
[0060] The method for calculating the comprehensive text output value is to establish a comprehensive data analysis model, import any text output value into the comprehensive data analysis model, and calculate the comprehensive text output value, which is specifically expressed as:
[0061] ,
[0062] φ represents the comprehensive text output value of the target large computing power model optimization system, φ1' represents the standard value of the model structure optimization evaluation value of the target large computing power model optimization system, and φ2' represents the standard value of the model process optimization evaluation value of the target large computing power model optimization system.
[0063] The human-computer interaction module outputs information based on the comprehensive text output value and the user information terminal, determines the optimization capability of the large computing power model optimization system, and generates a test report.
[0064] The method for judging the optimization capability of the large computing power model optimization system is to establish a standard value for the comprehensive text output value and mark it as φ 标 , when φ<φ 标 When φ≥φ 标 , it indicates that the comprehensive text output value is greater than or equal to the standard value of the comprehensive text output value, and the optimization capability of the target large computing power model optimization system meets the standard.
[0065] What needs to be specifically explained in this embodiment is that Jmeter is used to generate a test report, and the first key text collected by the above-mentioned optimization data collection module is imported into the tool and database for generating the test report, and the second key text, arbitrary text output value and comprehensive text output value obtained by the above-mentioned optimization data preprocessing module, optimization data analysis module and comprehensive data analysis module are imported into the tool and database for generating the test report. It will automatically generate a test report, store various parameters and evaluation values in the system testing process in the database, and feed back the test results to the target large computing power model optimization system, which will improve the optimization plan for the large computing power model according to the judgment results.
[0066] Secondly: In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other;
[0067] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A large computing power model optimization system based on artificial intelligence, including a system operation database, a system central processing module and a user information terminal, characterized in that: It also includes a running times determination module, an optimization data acquisition module, an optimization data preprocessing module, an optimization data analysis module, a comprehensive data analysis module and a human-computer interaction module; The system operation database includes all data texts of the large computing power model optimization system, and collects the information texts output by each module in real time. The system central processing module is used to centrally control the information text instructions output by each module, and the user information terminal is an information output device for receiving the large computing power model optimization system; The operation number determination module is used to filter out characteristic text data from the data text, and the number of times of filtering is recorded as 1, 2...n in sequence; The optimized data collection module selects the first key text based on the characteristic text data; The optimized data preprocessing module is used to perform preprocessing analysis on the first key text to obtain the second key text; The optimization data analysis module obtains any text output value in the operation number determination module based on the second key text analysis; The comprehensive data analysis module obtains a comprehensive text output value based on any text output value in the optimized data analysis module; The human-computer interaction module outputs information based on the comprehensive text output value and the user information terminal, determines the optimization capability of the large computing power model optimization system, and generates a test report; The first key text includes calculation cost parameters, model stability parameters, training process optimization parameters and inference process optimization parameters. The calculation cost parameters include the number of model layers, model dimensions, number of shared parameters, number of quantization parameters and data compression ratio, which are marked as L M , D M 、N S 、N Q and R D ; Model stability parameters include signal-to-noise ratio, model outliers, number of training data, and model complexity, which are marked as R S 、M O 、N T and C M ; The optimization parameters of the training process include the degree of reduction in training accuracy, the number of parallel nodes added, and the number of gradient checkpoints, which are marked as R A 、N P and N C ; The optimization parameters of the inference process include the number of key-value pairs stored, the degree of hardware acceleration, and the number of pruned neurons, which are marked as N K , D H and N E ; The second key text is obtained by constructing an optimized data preprocessing model, importing the collected computing cost parameters, model stability parameters, training process optimization parameters, and reasoning process optimization parameters into the optimized data preprocessing model, and calculating the second key text, including a computing cost evaluation value, a model stability evaluation value, a training process optimization evaluation value, and a reasoning process optimization evaluation value. The computing cost evaluation value is specifically expressed as: , A C represents the computational cost evaluation value of the target large computing power model optimization system, L M represents the number of model layers of the target large computing power model optimization system, D M N represents the model dimension of the target large computing power model optimization system. S Represents the number of shared parameters of the target large computing model optimization system, N Q Represents the number of quantitative parameters of the target large computing model optimization system, R Di represents the data compression ratio of the target large computing power model optimization system during the i-th operation, δ1 and δ2 represent the influence coefficients of the number of model layers and model dimensions on the computational cost evaluation value, respectively, and b1 and b2 are known constants; the model stability evaluation value is specifically expressed as: , A S represents the model stability evaluation value of the target large computing power model optimization system, R Si represents the signal-to-noise ratio of the target large computing power model optimization system during the i-th operation, M O Represents the model outlier value of the target large computing power model optimization system, N Ti represents the amount of training data during the i-th operation of the target large computing power model optimization system, C M represents the model complexity of the target large computing power model optimization system, δ3 represents the influence coefficient of model complexity on the model stability evaluation value, and b3 is a known constant; the optimization evaluation value of the training process is specifically expressed as: , A T represents the training process optimization evaluation value of the target large computing power model optimization system, R Ai Indicates the degree of reduction in training accuracy during the i-th operation of the target large computing power model optimization system, N Pi N represents the number of parallel nodes added during the i-th operation of the target large computing power model optimization system. Ci represents the number of gradient checkpoints during the i-th operation of the target large computing power model optimization system, and C1 represents a known constant; the optimization evaluation value of the inference process is specifically expressed as: , A R represents the optimization evaluation value of the inference process of the target large computing power model optimization system, N Ki represents the number of key-value pairs stored during the i-th operation of the target large computing power model optimization system, D H Indicates the hardware acceleration level of the target large computing model optimization system, N Ei It represents the number of pruned neurons during the i-th operation of the target large computing power model optimization system, and C2 and C3 represent known constants.
2. The artificial intelligence-based large computing power model optimization system according to claim 1, characterized in that: The method for filtering out characteristic text data is to monitor the operating status of the large computing power model system in real time, and use a log tool to record data during each system operation, and collect characteristic text data during the operation of the large computing power model optimization system through characteristic word screening.
3. The artificial intelligence-based large computing power model optimization system according to claim 1, characterized in that: The user information terminal is bound to a mobile phone or any electronic device capable of output based on the administrator's work information.
4. The artificial intelligence-based large computing power model optimization system according to claim 1, characterized in that: The method for calculating the arbitrary text output value is to establish an optimized data analysis model, and import the second key text into the optimized data analysis model to calculate the arbitrary text output value, including the model structure optimization evaluation value and the model process optimization evaluation value. The model structure optimization evaluation value is specifically expressed as: ,φ1 represents the model structure optimization evaluation value of the target large computing power model optimization system; The model process optimization evaluation value is specifically expressed as: , φ2 represents the model process optimization evaluation value of the target large computing power model optimization system.
5. The artificial intelligence-based large computing power model optimization system according to claim 1, characterized in that: The method for calculating the comprehensive text output value is to establish a comprehensive data analysis model, import any text output value into the comprehensive data analysis model, and calculate the comprehensive text output value, which is specifically expressed as: , φ represents the comprehensive text output value of the target large computing power model optimization system, φ1' represents the standard value of the model structure optimization evaluation value of the target large computing power model optimization system, and φ2' represents the standard value of the model process optimization evaluation value of the target large computing power model optimization system.
6. The artificial intelligence-based large computing power model optimization system according to claim 1, characterized in that: The method for judging the optimization capability of the large computing power model optimization system is to establish a standard value for the comprehensive text output value and mark it as φ 标 , when φ<φ 标 When φ≥φ 标 , it indicates that the comprehensive text output value is greater than or equal to the standard value of the comprehensive text output value, and the optimization capability of the target large computing power model optimization system meets the standard.
Citation Information
Patent Citations
Front-end text analysis method based on multi-task learning
CN118332121A
Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine
WO2020000248A1
