Base model optimization strategy determination method, device, medium, and product
By acquiring the task type labels and target indicator data combination of downstream tasks, the optimization problems of the base model are identified and optimization strategy combinations are generated, which solves the problem of low optimization efficiency of the base model in the existing technology and improves accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI YINGMIAO INTELLIGENT TECHNOLOGY CO LTD
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-05
AI Technical Summary
The existing base model has low optimization efficiency, mainly due to the lack of a systematic optimization strategy and reliance on human experience, which leads to inaccurate optimization.
By acquiring the task type labels and target indicator data combinations of downstream tasks, the current indicator data combination of the base model is determined, key lifecycle data is obtained, problems to be optimized are identified based on this data, and a combination of targeted optimization strategies is generated, displaying an optimization suggestion report.
This improved the accuracy and efficiency of base model optimization, ensuring the relevance and accuracy of the optimization strategy, and enhancing the model optimization effect and efficiency.
Smart Images
Figure CN122155507A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a method, device, medium and product for determining a base model optimization strategy. Background Technology
[0002] With the popularization of pedestal model technology, scenarios for adapting pedestal models to downstream tasks are increasing. To better adapt pedestal models to downstream tasks, secondary optimization is usually required, including but not limited to secondary training. In existing technologies, pedestal models are typically optimized by algorithm engineers based on personal experience and training sample sets, which easily leads to low optimization efficiency. Summary of the Invention
[0003] This invention provides a method, device, medium, and product for determining a base model optimization strategy to solve the problem of low efficiency in existing base model optimization.
[0004] According to one aspect of the present invention, a method for determining an optimization strategy for a base model is provided, the method comprising: Obtain a combination of task type tags and target indicator data for downstream tasks, wherein the target indicator data combination includes target data for each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators; Determine the current indicator data combination of the base model for the downstream task, wherein the current indicator data combination includes the current data of each of the multiple predetermined evaluation indicators; If the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, the current lifecycle key data of the base model is obtained. The lifecycle key data includes hyperparameter configuration data, network structure parameters and inference logs. Based on the key lifecycle data, identify the optimization problems existing in the base model, and generate an optimization strategy combination for the current optimization round based on the task type label and the optimization problems. The report presents optimization recommendations, including the problem to be optimized and the combination of optimization strategies.
[0005] According to another aspect of the present invention, a base model optimization strategy determination apparatus is provided, the apparatus comprising: The first acquisition module is used to acquire a combination of task type tags and target indicator data for downstream tasks. The target indicator data combination includes target data for each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators. The indicator data module is used to determine the current indicator data combination of the base model for the downstream task. The current indicator data combination includes the current data of each of the multiple predetermined evaluation indicators. The second acquisition module is used to acquire the current life cycle key data of the base model when the current indicator data combination does not meet the target conditions corresponding to the target indicator data combination. The life cycle key data includes hyperparameter configuration data, network structure parameters and inference logs. The strategy generation module is used to determine the optimization problems existing in the base model based on the key life cycle data, and to generate the optimization strategy combination for the current optimization round based on the task type label and the optimization problems. The report generation module is used to display an optimization suggestion report that includes the problem to be optimized and the combination of optimization strategies.
[0006] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: One or more processors; Storage device for storing one or more programs. When one or more programs are executed by one or more processors, the one or more processors implement the base model optimization strategy determination method as described in any embodiment of the present invention.
[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the base model optimization strategy determination method according to any embodiment of the present invention.
[0008] According to another aspect of the present invention, a computer program product is provided, which, when executed by a processor, implements the base model optimization strategy determination method as described in any embodiment of the present invention.
[0009] The technical solution provided by this invention indicates that if the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, it means that the performance of the current base model has not reached expectations. Therefore, by acquiring the current lifecycle key data of the base model and determining the optimization problems existing in the base model based on the current lifecycle key data, the accuracy of the determination of the optimization problems can be guaranteed. Based on the optimization problems and the task type tags of downstream tasks, the optimization strategy combination for the current optimization round is generated, which can ensure that the optimization strategy combination is determined in a targeted and accurate manner, thereby ensuring the accuracy of the optimization suggestion report including the optimization strategy combination and improving the accuracy of the base model optimization based on the optimization suggestion report. This achieves the technical effects of preset target guidance, automatic location of optimization problems, and intelligent matching of optimization strategies, thus addressing the problem of low optimization efficiency caused by inaccurate matching of optimization strategies and significantly improving the model optimization efficiency and model optimization effect.
[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A flowchart illustrating the method for determining the base model optimization strategy provided in an embodiment of the present invention; Figure 2 This is another flowchart illustrating the method for determining the base model optimization strategy provided in an embodiment of the present invention; Figure 3A This is a schematic diagram of the structure of the base model optimization strategy determination device provided in an embodiment of the present invention; Figure 3B This is another structural schematic diagram of the base model optimization strategy determination device provided in an embodiment of the present invention; Figure 3C This is another structural schematic diagram of the base model optimization strategy determination device provided in an embodiment of the present invention; Figure 3D This is another structural schematic diagram of the base model optimization strategy determination device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0015] Figure 1 This is a flowchart illustrating the method for determining a base model optimization strategy according to an embodiment of the present invention. This embodiment is applicable to base model optimization. The method can be executed by a base model optimization strategy determination device, which can be implemented in hardware and / or software and can be configured in electronic devices such as computers or servers. Figure 1 As shown, the method in this embodiment includes: S110. Obtain the task type label and target indicator data combination of the downstream task. The target indicator data combination includes the target data of each of the multiple predetermined evaluation indicators.
[0016] Downstream tasks refer to real-world applications that require fine-tuning based on the base model. The base model is a general-purpose model trained on large-scale, diverse datasets. It possesses powerful fundamental capabilities but is not specifically designed for any particular task; that is, the base model and downstream tasks are not naturally compatible.
[0017] Task type labels are used to characterize the type of downstream tasks, such as classification tasks, detection tasks, and generation tasks.
[0018] Multiple predetermined evaluation metrics include, but are not limited to, prediction accuracy, inference latency, and training memory cost. For example, a combination of target metrics might include a model prediction accuracy greater than or equal to 95%, an inference latency less than or equal to 50 milliseconds, and training memory cost less than or equal to 100 GPUs.
[0019] In one embodiment, a user can directly input a combination of target indicator data through an interactive interface, or input the same combination based on a predetermined data input template. The predetermined data input templates differ for different task type labels. Each predetermined data input template includes the data ranges for all predetermined evaluation indicators. Thus, when the processor detects the user-inputted combination of target indicator data, it can directly verify the accuracy of the input data for each predetermined evaluation indicator based on the input data and its corresponding data range. For example, if the prediction accuracy rate of a predetermined evaluation indicator is 90%-99%, then if the input data for that predetermined evaluation indicator is greater than or equal to 99%, a corresponding prompt message will be output. This prompt message indicates that the input data is unreasonable and exceeds the highest prediction level of the base model.
[0020] S120. Determine the current indicator data combination of the base model for the downstream task. The current indicator data combination includes the current data of each of the multiple predetermined evaluation indicators.
[0021] After one optimization round, the current data for each of the multiple predetermined evaluation indicators of the current base model is obtained to obtain the current indicator data combination. This current indicator data combination can reflect the accuracy and stability of the base model's prediction results for downstream tasks.
[0022] S130. If the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, obtain the current lifecycle key data of the base model. The lifecycle key data includes hyperparameter configuration data, network structure parameters and inference logs.
[0023] If the performance of the downstream task corresponding to any current data in the current indicator data combination is lower than the performance of the downstream task corresponding to the target data, it means that the training result of the downstream task has not met expectations, and further optimization is needed. The key data of the current life cycle of the base model needs to be automatically obtained through the model performance evaluation module.
[0024] For example, the target data for model accuracy is greater than or equal to 95%. If its current data is 90%, it is determined that the current base model's prediction accuracy for downstream tasks is lower than the expected prediction accuracy. The target data for inference latency is less than or equal to 50 milliseconds. If its current data is 60 milliseconds, it is determined that the current base model's inference latency for downstream tasks is lower than the expected inference latency.
[0025] Hyperparameter configuration data belongs to the configuration metadata of the pre-training phase, guiding the training of the base model, but these are settings that do not participate in gradient updates. Network structure parameters are generated upon completion of training and exist as the carrier of model capabilities. Inference logs are operational data after the base model is deployed, i.e., recorded data generated during model service, which can be used as a basis for monitoring and optimization. Of course, key lifecycle data may also include training data distribution, cleaning rules, weights, etc. In actual use, these can be set according to the actual situation.
[0026] S140. Based on key lifecycle data, identify the optimization problems existing in the base model, and based on task type labels and optimization problems, generate the optimization strategy combination for the current optimization round.
[0027] Once the key lifecycle data is determined, feature engineering is used to extract the core optimization problems and classify and label the problem types, such as the first problem for the data, the second problem for the model, and the third problem for the parameters.
[0028] For example, the first problem with the data includes, but is not limited to, at least one of the following: imbalanced sample size and high noise level; the second problem with the model includes, but is not limited to, at least one of the following: overfitting and insufficient network parameters; and the third problem with the parameters includes, but is not limited to, at least one of the following: inaccurate learning rate and regularization strength below a threshold.
[0029] In one embodiment, the key lifecycle data is input into a pre-trained problem identification model to obtain the problem to be optimized. This embodiment can quickly and accurately identify the problem to be optimized with high accuracy.
[0030] After determining the task type and the problem to be optimized, both are used as matching targets. The matching degree between each strategy in the strategy library and the matching target is determined, and strategies with a matching degree greater than a predetermined matching threshold are used as target strategies for the matching target. The combination of all target strategies is used as the optimization strategy combination. In this embodiment, the matching degree between each strategy in the strategy library and the matching target can be determined based on a similarity matching algorithm or based on a pre-trained matching model.
[0031] In one embodiment, the optimization strategy combination includes at least one of data augmentation strategy, network structure adjustment strategy and hyperparameter tuning strategy, as well as adaptation scenario labels and effect quantification data associated with each of the optimization strategies.
[0032] Data augmentation strategies are a set of methods that improve the generalization ability, robustness, and performance of models by meaningfully transforming and expanding existing training data to generate new and diverse training samples.
[0033] Network structure adjustment strategies are a set of methods that systematically modify the neural network architecture, connection methods, or parameter organization without changing the core functionality of the model, in order to optimize the model's performance, efficiency, generalization ability, or specific characteristics.
[0034] Hyperparameter tuning strategies are a set of methods that systematically search, evaluate, and select fixed configuration parameters during model training to maximize model performance, efficiency, or stability. These parameters do not participate in gradient updates but control various aspects of the learning process.
[0035] Scene tags are a structured, standardized, and multi-dimensional metadata system used to accurately describe the complete context of a machine learning task or model application, including all relevant factors such as problem characteristics, data conditions, resource constraints, business requirements, and technical environment.
[0036] Performance quantification data is a numerical evidence system that systematically measures, evaluates, and compares the effectiveness of different optimization strategies in specific scenarios. It includes, but is not limited to, baseline performance data, optimization benefit data, cost data, stability data, and robustness data.
[0037] S150. Display an optimization suggestion report that includes the problem to be optimized and the combination of optimization strategies.
[0038] In one embodiment, the optimization suggestion report only includes the problem to be optimized and the combination of optimization strategies, which presents the combination of optimization strategies to the user in a simple and clear manner.
[0039] In one embodiment, the optimization suggestion report includes the problem to be optimized, the combination of optimization strategies, quantitative data on the effects of each optimization strategy in the combination, and their priorities. This structurally integrates the problem description information corresponding to the problem to be optimized, the combination of optimization strategies corresponding to the problem to be optimized, the expected effects that the combination of optimization strategies can achieve, and the priorities of each optimization strategy in the combination into the optimization suggestion report. This optimization suggestion report can be output in text format or displayed visually. Thus, through this optimization suggestion report, users can intuitively and accurately understand which optimization strategies need to be used to optimize the base model in the current optimization round, and the optimization effects corresponding to each optimization strategy.
[0040] In one embodiment, an optimization suggestion template is received, the optimization strategies are combined and integrated into the optimization suggestion template, an optimization suggestion report is generated, and the optimization suggestion report is displayed. This embodiment provides a personalized optimization suggestion report generation method that can meet the needs of different users.
[0041] The technical solution provided by this invention indicates that if the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, it means that the performance of the current base model has not reached expectations. Therefore, by acquiring the current lifecycle key data of the base model and determining the optimization problems existing in the base model based on the current lifecycle key data, the accuracy of the determination of the optimization problems can be guaranteed. Based on the optimization problems and the task type tags of downstream tasks, the optimization strategy combination for the current optimization round is generated, which can ensure that the optimization strategy combination is determined in a targeted and accurate manner, thereby ensuring the accuracy of the optimization suggestion report including the optimization strategy combination and improving the accuracy of the base model optimization based on the optimization suggestion report. This achieves the technical effects of preset target guidance, automatic location of optimization problems, and intelligent matching of optimization strategies, thus addressing the problem of low optimization efficiency caused by inaccurate matching of optimization strategies and significantly improving the model optimization efficiency and model optimization effect.
[0042] Building upon the aforementioned embodiments, while displaying the optimization suggestion report, the base model is automatically optimized based on the optimization strategy combination in the report; or, upon detecting strategy implementation information, the base model is optimized according to the optimization strategy combination in the optimization suggestion report; or, upon receiving implementation guidance information from the optimization suggestion report, the base model is optimized according to the optimization strategy corresponding to the implementation guidance information, wherein each optimization strategy corresponds to one implementation guidance information. This embodiment progressively completes the base model optimization operation corresponding to each optimization strategy in the optimization strategy combination through interactive information regarding strategy implementation.
[0043] Based on the aforementioned embodiments, when the feedback time is detected to have arrived, the implementation progress information of the optimized strategy combination is displayed.
[0044] Specifically, the implementation progress display interval is a modifiable item that users can determine themselves. When optimizing the base model based on the optimization strategies included in the optimization suggestion report, the implementation progress of each optimization strategy in the current iterative optimization strategy combination will be displayed when the implementation progress display interval is reached. This allows users to understand the implementation progress of each optimization strategy in a timely manner.
[0045] Based on the aforementioned embodiments, when the implementation of the optimization strategy combination is detected to have ended, the step of determining the current indicator data combination of the base model for the downstream task is returned until the current indicator data combination meets the target conditions corresponding to the target indicator data combination, a target model is generated, and optimization end information is output.
[0046] Specifically, the completion of the optimization strategy combination implementation signifies the end of the corresponding optimization round. It's necessary to determine whether the optimization results meet expectations. Therefore, the process returns to step S120, which determines the current combination of indicator data for the base model in relation to the downstream task. Steps S120-S150 are repeated until the current combination of indicator data meets the target conditions corresponding to the target combination of indicator data. A compliance report is then output, concluding the base model optimization process for the downstream task. This embodiment can automatically generate an optimization suggestion report and complete multiple rounds of base model optimization based on this report.
[0047] Based on the aforementioned embodiments, when the implementation of the optimization strategy combination is detected to have ended, the current indicator data combination for the current base model is determined, and the current indicator data combination is compared with the indicator data combination under the previous optimization round to obtain the optimization contribution data of the current optimization round. The optimization contribution data includes the change data of each predetermined evaluation indicator. A result report is generated based on the optimization contribution data. The result report is used to show the contribution of the current iteration round to each predetermined evaluation indicator.
[0048] Based on the foregoing embodiments, upon receiving feedback information regarding the optimization strategy combination, the optimization strategy combination is updated according to the feedback information, and the optimization suggestion report is updated according to the updated optimization strategy combination.
[0049] Specifically, after the optimization suggestion report is generated, the user can input feedback information on the exchange interface. Upon detecting this feedback, the processor extracts target information from it. This information includes at least one of the following: custom strategy requirements, the acceptability of some or all strategies in the optimized strategy combination, and the expected score of the implementation effect of some or all strategies in the optimized strategy combination. Then, based on the target information, the processor adjusts the weights of strategies in a pre-created strategy library to generate optimization requirements for the feedback. Based on these requirements, optimization strategies are re-matched from the strategy library. This embodiment, through human-computer interaction, obtains user feedback information on the optimized strategy combination, updates the optimized strategy combination based on this feedback, and updates the optimization suggestion report based on the updated optimized strategy combination. This improves the accuracy of the optimization suggestion report and achieves the technical effects of preset goal guidance, automatic identification of problems to be optimized, intelligent matching of optimization strategies, and feedback closed-loop adjustment. It solves the defects of low model optimization efficiency, inaccurate strategy matching, and lack of process controllability in existing technologies, lowers the threshold for model optimization, and improves optimization results and user experience.
[0050] Regarding feedback input: Users can directly input feedback through the interactive interface or input feedback based on a pre-defined template.
[0051] Figure 2This is a flowchart illustrating the method for determining the base model optimization strategy provided in an embodiment of the present invention. Figure 2 As shown, the method in this embodiment may specifically include: First, the target indicator data combination is set through the target setting module. The current indicator data combination of the base model in the current optimization round is determined. The performance of the downstream task corresponding to each indicator data in the current indicator data combination is then checked against the performance of the downstream task corresponding to the target data. If the performance is lower, the model has achieved its target, an achievement report is output, and the process ends. If not, the current lifecycle key data of the base model is obtained. Based on this lifecycle key data, optimization problems in the base model are identified. Based on the task type label and the optimization problems, highly suitable optimization strategies are detected from the strategy library, generating the optimization strategy combination for the current optimization round. A structured optimization suggestion report is then generated. The optimization suggestion report includes: combining optimization strategies; pushing the optimization suggestion report and simultaneously initiating optimization monitoring operations to monitor the optimization operations on the base model; receiving user feedback and adjusting the optimization strategy combination based on the feedback; optimizing the base model based on the adjusted optimization strategy combination; after the current iteration optimization is completed, returning to the step of determining the current indicator data combination of the downstream task in the current optimization round, until the performance of the downstream task corresponding to each current data in the current indicator data combination is not lower than the performance of the downstream task corresponding to the corresponding target data, generating a base model compliance report, at which point the base model optimization is completed and becomes the target model that meets the prediction requirements of the downstream task.
[0052] Figure 3A This is a schematic diagram of the base model optimization strategy determination device provided in an embodiment of the present invention. Figure 3A As shown, the base model optimization strategy determination device includes: The first acquisition module 110 is used to acquire a combination of task type tags and target indicator data of downstream tasks, wherein the target indicator data combination includes target data of each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators. The indicator data module 120 is used to determine the current indicator data combination of the base model for the downstream task, wherein the current indicator data combination includes the current data of each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators; The second acquisition module 130 is used to acquire the current life cycle key data of the base model when the current indicator data combination does not meet the target conditions corresponding to the target indicator data combination. The life cycle key data includes hyperparameter configuration data, network structure parameters and inference logs. The strategy generation module 140 is used to determine the optimization problems existing in the base model based on the key life cycle data, and to generate the optimization strategy combination for the current optimization round based on the task type label and the optimization problems. The report generation module 150 is used to display an optimization suggestion report that includes the problem to be optimized and the combination of optimization strategies.
[0053] The technical solution provided by this invention indicates that if the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, it means that the performance of the current base model has not reached expectations. Therefore, by acquiring the current lifecycle key data of the base model and determining the optimization problems existing in the base model based on the current lifecycle key data, the accuracy of the determination of the optimization problems can be guaranteed. Based on the optimization problems and the task type tags of downstream tasks, the optimization strategy combination for the current optimization round is generated, which can ensure that the optimization strategy combination is determined in a targeted and accurate manner, thereby ensuring the accuracy of the optimization suggestion report including the optimization strategy combination and improving the accuracy of the base model optimization based on the optimization suggestion report. This achieves the technical effects of preset target guidance, automatic location of optimization problems, and intelligent matching of optimization strategies, thus addressing the problem of low optimization efficiency caused by inaccurate matching of optimization strategies and significantly improving the model optimization efficiency and model optimization effect.
[0054] In one embodiment, the plurality of predetermined evaluation metrics include model prediction accuracy, inference latency, and training memory cost; The problems to be optimized include a first problem concerning the data, a second problem concerning the model, and a third problem concerning the parameters.
[0055] In one embodiment, the optimization strategy combination includes at least one of data augmentation strategy, network structure adjustment strategy and hyperparameter tuning strategy, as well as adaptation scenario labels and effect quantification data corresponding to each optimization strategy.
[0056] In one embodiment, the optimization suggestion report includes the problem to be optimized, the combination of optimization strategies, quantitative data on the effects of each optimization strategy in the combination of optimization strategies, and their priorities.
[0057] In one embodiment, such as Figure 3B As shown, the device also includes a feedback optimization module 160, which is used for: Upon receiving feedback information regarding the optimized strategy combination, the optimized strategy combination and the optimized suggestion report are updated based on the feedback information.
[0058] In one embodiment, such as Figure 3CAs shown, the device also includes a progress display module 170, which is used for: If the feedback time is detected, the implementation progress information of the optimized strategy combination is displayed.
[0059] In one embodiment, such as Figure 3D As shown, the module also includes an iterative optimization module 180, which is used for: If the implementation of the optimization strategy combination is detected to be completed, the step of determining the current indicator data combination of the base model for the downstream task is returned until the current indicator data combination meets the target conditions corresponding to the target indicator data combination, the target model is generated, and the optimization completion information is output.
[0060] The base model optimization strategy determination device provided in the embodiments of the present invention can execute the base model optimization strategy determination method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.
[0061] It is worth noting that the various units and modules included in the above-mentioned base model optimization strategy determination device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of the present invention.
[0062] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.
[0063] like Figure 4As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0064] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0065] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the base model optimization strategy determination method.
[0066] In some embodiments, the base model optimization strategy determination method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via read-only memory (ROM) 12 and / or communication unit 19. When the computer program is loaded into random access memory (RAM) 13 and executed by processor 11, one or more steps of the base model optimization strategy determination method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the base model optimization strategy determination method by any other suitable means (e.g., by means of firmware).
[0067] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0068] Computer programs used to implement the base model optimization strategy determination method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.
[0069] This invention provides a computer-readable storage medium storing computer instructions for causing a processor to execute a base model optimization strategy determination method, including: Obtain a combination of task type tags and target indicator data for downstream tasks, wherein the target indicator data combination includes target data for each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators; Determine the current indicator data combination of the base model for the downstream task, wherein the current indicator data combination includes the current data of each of the multiple predetermined evaluation indicators; If the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, the current lifecycle key data of the base model is obtained. The lifecycle key data includes hyperparameter configuration data, network structure parameters and inference logs. Based on the key lifecycle data, identify the optimization problems existing in the base model, and generate an optimization strategy combination for the current optimization round based on the task type label and the optimization problems. The report presents optimization recommendations, including the problem to be optimized and the combination of optimization strategies.
[0070] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0071] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0072] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.
[0073] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.
[0074] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.
[0075] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements a base model optimization strategy determination method according to any embodiment of the invention.
[0076] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof. Programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0077] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0078] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A method for determining the optimization strategy of a base model, characterized in that, The method includes: Obtain a combination of task type tags and target indicator data for downstream tasks, wherein the target indicator data combination includes target data for each of the predetermined evaluation indicators among a plurality of predetermined evaluation indicators; Determine the current indicator data combination of the base model for the downstream task, wherein the current indicator data combination includes the current data of each of the multiple predetermined evaluation indicators; If the current combination of indicator data does not meet the target conditions corresponding to the target combination of indicator data, the current lifecycle key data of the base model is obtained. The lifecycle key data includes hyperparameter configuration data, network structure parameters and inference logs. Based on the key lifecycle data, identify the optimization problems existing in the base model, and generate an optimization strategy combination for the current optimization round based on the task type label and the optimization problems. The report presents optimization recommendations, including the problem to be optimized and the combination of optimization strategies.
2. The method according to claim 1, characterized in that, The predetermined evaluation metrics include model prediction accuracy, inference latency, and training memory cost. The problems to be optimized include a first problem concerning the data, a second problem concerning the model, and a third problem concerning the parameters.
3. The method according to claim 1, characterized in that, The optimization strategy combination includes at least one of data augmentation strategy, network structure adjustment strategy and hyperparameter tuning strategy, as well as the adaptation scenario label and effect quantification data corresponding to each optimization strategy.
4. The method according to claim 1, characterized in that, The optimization suggestion report includes the problem to be optimized, the combination of optimization strategies, and the quantitative data and priority of the effect of each optimization strategy in the combination of optimization strategies.
5. The method according to claim 1, characterized in that, Also includes: Upon receiving feedback information regarding the optimized strategy combination, the optimized strategy combination and the optimized suggestion report are updated based on the feedback information.
6. The method according to claim 1, characterized in that, Also includes: If the feedback time is detected, the implementation progress information of the optimized strategy combination is displayed.
7. The method according to claim 1, characterized in that, Also includes: If the implementation of the optimization strategy combination is detected to be completed, the step of determining the current indicator data combination of the base model for the downstream task is returned until the current indicator data combination meets the target conditions corresponding to the target indicator data combination, the target model is generated, and the optimization completion information is output.
8. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the base model optimization strategy determination method as described in any one of claims 1-7.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the base model optimization strategy determination method according to any one of claims 1-7.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for determining the base model optimization strategy according to any one of claims 1-7.