Training task data processing method, device and equipment
By using a cascaded structure of large language models and lightweight professional models, combined with deep semantic analysis and error feedback mechanisms, the inefficiency of complex reasoning and resource allocation in simulated attack training is solved, achieving efficient and accurate execution of training tasks.
Patent Information
- Application Number
- CN202411003270.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-08
AI Technical Summary
Existing technologies cannot efficiently handle complex reasoning and resource allocation tasks in simulated attack training, especially the understanding and accurate calculation of unstructured natural language information, resulting in low efficiency and susceptibility to errors.
By combining a cascaded structure of large language models and lightweight professional models, and through deep semantic analysis and error feedback mechanisms, efficient processing of training task data is achieved.
It improves the accuracy and efficiency of training tasks, ensures smooth task execution, and simplifies the parsing and computation process of unstructured data.
Smart Images

Figure CN118917438A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of computer technology, and specifically relates to a training task data processing method, device and equipment. Background Art
[0002] For resource allocation problems described in natural language, especially in the process of simulated attack training, such as "In a certain training mission, the enemy set up three strongholds outside our base, and their threat levels are (2, 3, 1). We plan to send a bombardment force composed of ten fire units to destroy the enemy strongholds. According to reconnaissance intelligence and terrain analysis, there are three combat areas that can be used as positions to deploy our attack units. The probability of each position destroying each enemy stronghold is: the first position (0.29, 0.76, 0.55), the second position (0.56, 0.39, 0.57), and the third position (0.45, 0.41, 0.73). Now we need to plan the number of attack units deployed in each training area and determine the optimal attack strategy for each attack point to maximize the destruction of the enemy stronghold." The task in the example cannot be directly handled by the traditional attack resource allocation or position deployment model. It requires manual analysis and extraction of data, and then input into the algorithm model for calculation. This is not only inefficient, but also prone to errors. With the development of technology, the sources of training instructions and intelligence information are becoming increasingly diversified, including a large amount of unstructured natural language information, which poses new challenges to the rapid understanding of information and the efficient solution of small models.
[0003] Artificial intelligence technology, especially breakthroughs in natural language processing, provides new methods to solve the above problems. Large language models (LLMs) have the potential to assist decision-making. In the field of simulated attack training, LLMs have been used for information analysis, situational awareness, and decision support on the training ground. For example, LLMs can help commanders quickly understand and process a large amount of training ground information and provide reasonable decision-making suggestions. LLMs have made significant progress in the field of natural language processing, showing strong context understanding, dialogue generation, and abstract reasoning capabilities. Causal reasoning methods, knowledge-guided reasoning, and prompt engineering methods can use LLMs to deal with complex logical reasoning problems. These studies bring broader space for the implementation of LLMs. However, it is currently impossible to use LLMs to directly handle relatively complex reasoning and cognitive tasks, especially in highly specialized and computationally intensive tasks such as precise attack resource allocation within the scope of training actions for attack matters. Although LLMs can assist in understanding complex training instructions and environmental descriptions, to deal with precise calculations and decision optimization problems such as attack resource allocation, they need to be combined with small model algorithms with professional computing capabilities. Therefore, how to effectively combine the advanced cognitive capabilities of LLMs with the efficient computing advantages of small models in specific tasks has become the key to improving the level of command automation and attack training effectiveness on the training ground. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a training task data processing method, device and equipment that can process training task data more accurately and efficiently to determine whether the training task can be successfully executed.
[0005] The present invention includes a training task data processing method, comprising:
[0006] In response to the obtained training task data, performing deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language;
[0007] Determine a target task type that satisfies a mapping condition with the training task based on the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task;
[0008] Determining a candidate computing model based on the target task type;
[0009] Adjusting the candidate computing model based on the detailed information to generate a target computing model;
[0010] Calling the target computing model to try to execute the training task, completing the data calculation required for the task, and obtaining the execution result;
[0011] When the execution result indicates that the task execution has failed, an error feedback mechanism is executed to regenerate the target computing model in combination with the collected error information.
[0012] In some embodiments, the method further comprises:
[0013] The training tasks, training task types, and detailed information of the training tasks are adjusted based on a preset output format to form structured data, where the structured data is in a data format that can be recognized by the computing model.
[0014] In some embodiments, determining the target task type that satisfies the mapping condition with the training task based on the training task type includes:
[0015] Pre-build multiple different predefined task types based on different task types;
[0016] Based on the training task type, a target task type corresponding to the training task is determined by matching among a plurality of predefined task types constructed.
[0017] In some embodiments, determining a candidate computing model based on the target task type includes:
[0018] Based on the target task type, a corresponding computing model is determined as a candidate computing model, and a model code of the candidate computing model is determined.
[0019] In some embodiments, adjusting the candidate computing model based on the detail information to generate a target computing model includes:
[0020] Adjusting parameters of the model code of the candidate computing model based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information, so as to generate the target computing model based on the adjusted model code; or
[0021] Determine the required toolkit based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information;
[0022] The toolkit is imported into the model code of the candidate computing model to generate the target computing model based on the adjusted model code.
[0023] In some embodiments, each of the predefined task types has a corresponding example code involved in performing the task of that type;
[0024] The method further comprises:
[0025] The example code is adjusted based on the detailed information, and the example code is configured to form a target code that matches the training task and can call the target computing model.
[0026] In some embodiments, calling the target computing model to try to execute the training task includes:
[0027] The target code is executed based on a coding parser, and the target computing model is called to perform computing tasks during the execution of the target code. The computing tasks are different for different training tasks.
[0028] In some embodiments, when the execution result indicates that the task execution fails, executing the error feedback mechanism includes:
[0029] When the execution result indicates that the task has failed, error information is collected;
[0030] The target code and / or target computing model is modified based on the error information to form a new target code and / or target computing model.
[0031] Another embodiment of the present invention also provides a training task data processing device, including:
[0032] A response module, configured to respond to the obtained training task data and perform deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language;
[0033] A first determination module is used to determine a target task type that satisfies a mapping condition with the training task according to the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task;
[0034] A second determination module, used to determine a candidate computing model according to the target task type;
[0035] A first adjustment module, configured to adjust the candidate computing model according to the detail information to generate a target computing model;
[0036] A calling module is used to call the target computing model to try to execute the training task, complete the data calculation required for the task, and obtain the execution result;
[0037] The error feedback module is used to execute the error feedback mechanism when the execution result indicates that the task execution fails, so as to regenerate the target computing model in combination with the collected error information.
[0038] Another embodiment of the present invention further provides an electronic device, including:
[0039] at least one processor; and,
[0040] a memory communicatively connected to the at least one processor; wherein,
[0041] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to implement the training task data processing method as described in any one of the embodiments above.
[0042] The beneficial effects of the embodiments of the present invention include that the overall process of training task data processing includes four stages: task analysis, task planning, task execution, and error feedback. Based on the processing of each stage, the system can directly realize task planning and task execution based on the training task data edited in natural language, which simplifies the user's processing operations on the training task data edited in natural language. Moreover, because the matching computing model can be selected when executing the task in this scheme to realize the computing task of specific task data, the system not only has accurate data analysis capabilities and can realize the analysis of unstructured languages, but also has efficient and accurate computing capabilities. At the same time, through the subsequent error feedback mechanism, the task execution process can be continuously improved, which effectively ensures that the training tasks can be executed smoothly and effectively. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 Schematic diagram of the flow of the training task data processing method in an embodiment of the present invention.
[0045] Figure 2 Schematic diagram of the application flow of the training task data processing method in an embodiment of the present invention.
[0046] Figure 3 4 is a structural block diagram of a training task data processing device in an embodiment of the present invention. DETAILED DESCRIPTION
[0047] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, but are not intended to limit the present invention.
[0048] It should be understood that various modifications may be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope of the present disclosure will occur to those skilled in the art.
[0049] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.
[0050] These and other characteristics of the invention will become apparent from the following description of a preferred form of embodiment given as a non-limiting example, with reference to the accompanying drawings.
[0051] It should also be understood that, although the invention has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the invention that have the characteristics recited in the claims and are therefore within the scope of protection defined thereby.
[0052] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.
[0053] Specific embodiments of the present disclosure are described hereinafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of ways with substantially any suitable detailed structure.
[0054] This specification may use the phrases "in one embodiment," "in another embodiment," "in a further embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present disclosure.
[0055] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0056] As shown in the figure, the present invention includes a training task data processing method, including:
[0057] S1: In response to the obtained training task data, perform deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language;
[0058] S2: determining a target task type that satisfies a mapping condition with the training task based on the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task;
[0059] S3: Determine a candidate computing model based on the target task type;
[0060] S4: adjusting the candidate computing model based on the detailed information to generate a target computing model;
[0061] S5: calling the target computing model to try to execute the training task, completing the data calculation required for the task, and obtaining the execution result;
[0062] S6: When the execution result indicates that the task execution has failed, an error feedback mechanism is executed to regenerate a target computing model in combination with the collected error information.
[0063] Exemplarily, some steps in the scheme in this embodiment, especially the analysis of training task data, etc., can be but not limited to the large language model, and can be specifically modified by modifying the code of the large language model to enable it to execute the process steps in this embodiment. In practical application, the scheme of this embodiment can be considered to be a hierarchical cascade structure composed of an improved large language model and a lightweight professional model with high computing power, that is, a computing model, to jointly complete the processing of training task data, which is intended to overcome the limitations of a single model and integrate the advantages of large and small models of multiple intelligence. The scheme in this embodiment first uses the extensive knowledge base and deep understanding ability of the large language model to parse complex and unstructured natural language instructions (training task data), distinguish task types and form structured data. Subsequently, these data are accurately directed to computing models (lightweight professional models), which focus on specific field tasks, such as accurate firepower calculation, etc., and perform efficient and targeted processing. The computing model can quickly generate accurate solutions or suggestions by virtue of its optimization algorithm and fast computing characteristics in specific fields. During this process, the two models work closely together through intelligent routing and information refinement mechanisms to form a deep and broad collaborative reasoning framework, which not only retains the generalization and understanding advantages of large language models, but also integrates the computational efficiency and domain expertise of computational models.
[0064] Specifically, the solution of this embodiment mainly includes:
[0065] 1) Resource optimization and efficiency improvement of cascade structure: Through the cascade structure, knowledge-intensive tasks and computation-intensive tasks are assigned to the large language model and the computation model respectively, achieving efficient configuration of computing resources. The large language model is responsible for understanding and strategy formulation, reducing unnecessary computing burden; the computation model focuses on efficient execution and ensures rapid response, thereby improving the overall ability and efficiency of the model to handle complex tasks.
[0066] 2) Error feedback mechanism: Due to its characteristics, large language models may produce unexpected results or errors. In order to correct errors and improve accuracy, this embodiment designs an error feedback correction mechanism based on a code parser (such as a Python parser). Through this process, the large language model can adjust and correct collaborative reasoning errors by itself, produce more accurate operating results, and thus improve the accuracy and robustness of the cascade model.
[0067] 3) Flexibility and scalability: By building a modular collaborative reasoning framework with a large model as the core, different professional field computing models can be flexibly inserted or replaced according to task requirements, facilitating the continuous introduction of the latest AI research results, such as new large language models or computing models optimized for special tasks, and enhancing the generalization and scalability of collaborative reasoning.
[0068] Therefore, based on the above content, it can be known that the beneficial effects of the scheme of this embodiment include that the overall process of training task data processing includes four stages: task analysis, task planning, task execution, and error feedback. Based on the processing of each stage, the system can directly realize task planning and task execution based on the training task data edited in natural language, which simplifies the user's processing operations on the training task data edited in natural language. Moreover, because the matching computing model can be selected when executing the task in this scheme to realize the calculation task of specific task data, the system not only has accurate data analysis capabilities, can realize the analysis of unstructured language, but also has efficient and accurate computing capabilities. At the same time, through the subsequent error feedback mechanism, the task execution process can be continuously improved, which effectively ensures that the training tasks can be executed smoothly and effectively.
[0069] In one embodiment, the method further comprises:
[0070] S7: Adjust the training task, training task type, and detailed information of the training task based on a preset output format to form structured data, where the structured data is in a data format that can be recognized by the computing model.
[0071] Specifically, in the task analysis stage of this embodiment, since the training task data is unstructured data, the main problem faced in this stage is the identification and analysis of unstructured task problems. Figure 2 As shown, the training task data is Figure 2As mentioned in the record on the right side, traditional computing models can usually only process structured data and cannot process unstructured information such as natural language. In order to convert unstructured information into structured information, manual intervention is usually required, which is not only inefficient but also prone to errors. In order to solve this problem, the powerful natural language understanding ability of the large language model is used to analyze the task as a whole in this embodiment. Specifically, the large language model (hereinafter referred to as LLMs) first performs a deep semantic analysis of the task problem, and judges the task category, as well as the specific training tasks and detailed information of the training tasks through its huge knowledge base and semantic understanding ability, and adjusts and formats it according to the preset output format to form structured data, so that the subsequent computing model can recognize the structured data and complete the corresponding data calculation.
[0072] Further, the determining, based on the training task type, a target task type that satisfies a mapping condition with the training task includes:
[0073] S8: Pre-build multiple different predefined task types based on different task types;
[0074] S9: Based on the training task type, a target task type corresponding to the training task is determined by matching the multiple predefined task types constructed.
[0075] For example, LLMs accurately map task problems to predefined task categories, each of which corresponds to a specific processing method and strategy. In this process, LLMs can quickly and accurately understand the task type, improve the efficiency of task parsing, and provide more accurate solution directions and model support for task planning in subsequent stages.
[0076] The determining of the candidate computing model based on the target task type includes:
[0077] S10: Determine a corresponding computing model as a candidate computing model based on the target task type, and determine a model code of the candidate computing model.
[0078] The step of adjusting the candidate computing model based on the detail information to generate a target computing model includes:
[0079] S11: adjusting parameters of the model code of the candidate computing model based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information, so as to generate the target computing model based on the adjusted model code; or
[0080] S12: Determine the required toolkit based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information;
[0081] S13: Importing the toolkit into the model code of the candidate computing model to generate the target computing model based on the adjusted model code.
[0082] Moreover, in this embodiment, each of the predefined task types has a corresponding example code involved in executing the task of this type, that is, a code corresponding to the task.
[0083] The method further comprises:
[0084] S14: Adjust the example code based on the detailed information, and configure the example code to form a target code that matches the training task and can call the target computing model.
[0085] Specifically, in the task planning stage, LLMs identify the details and specific requirements of the task from the task problem based on the task type and given task information obtained in the task parsing stage, combined with the task examples and the specified output format. For example, LLMs can identify the number of firepower units involved in the training task, the deployment parameters and other related parameters to form structured data. This capability significantly reduces the need for manual processing of unstructured information, improves the efficiency and accuracy of task planning, and generates corresponding model codes. Specifically, LLMs will select the corresponding example code and model code according to the task type, and modify them in combination with the specific requirements of the task to generate a calculation model code suitable for a specific task and a target code for the corresponding training task. When adjusting to obtain the target calculation model code, when LLMs analyzes that the task is a firepower allocation type in the task parsing stage, you can choose to import the toolkit of the firepower allocation small model in the corresponding model code to form the target calculation model code, or you can continue to adjust based on the training task data after importing the toolkit, and then form structured data based on the information such as the number of firepower units extracted from the task problem, and finally solve the problem by calling the target calculation model. At this stage, LLMs not only need to understand the task information, but also need to make certain inferences and decisions to call the most appropriate computing model. LLMs also need to consider the executableness and optimization of the code when generating code to ensure the smooth progress of the subsequent task execution phase. In this way, the task planning stage realizes the structured processing of task data, ensures the accuracy and applicability of the generated code, and thus improves the speed and quality of task planning.
[0086] Further, calling the target computing model to try to execute the training task includes:
[0087] S15: Execute the target code based on the encoding parser, and call the target computing model to perform computing tasks during the execution of the target code. The computing tasks are different for different training tasks.
[0088] Wherein, when the execution result indicates that the task execution fails, executing the error feedback mechanism includes:
[0089] S16: When the execution result indicates that the task has failed, error information is collected;
[0090] S17: Modify the target code and / or target computing model based on the error information to form a new target code and / or target computing model.
[0091] Specifically, in this embodiment, in the task execution phase, the target code generated in the task planning phase is executed by the set Python code parser. During the code execution process, the selected calculation model will be called and execute a complex calculation process. Specifically, when the Python parser executes the model code generated by LLMs, the execution status of the code will be monitored in real time to ensure the smooth execution of the task. If the task is successfully executed, the system will output the calculation result; if the task fails, the error feedback mechanism will be triggered. The error feedback mechanism is an important part of the scheme of this embodiment. When the Python parser has an error when executing the model code generated by LLMs, the error feedback mechanism will collect error information, including error code, error information, task problem, task example, task information and output format. This information will be re-input to LLMs, and LLMs will modify the generated code according to this feedback information, regenerate more accurate target code and target model code, or only adjust one of them. Through this feedback loop, the task execution process is continuously optimized to improve the accuracy and reliability of the generated code. The task execution phase is not only the execution process of the code, but also includes the analysis and verification of the execution results. Through the task execution stage and error feedback mechanism, the method of this embodiment adopts the cascade structure of LLMs and computational models to separate the task understanding and computational process, significantly improving the efficiency and accuracy of task processing. The task execution and feedback mechanism at this stage ensures the robustness and reliability of the system and can cope with the challenges of various complex tasks.
[0092] For example, in a simulated battlefield, i.e., a real-life cos game, or when performing reconnaissance tasks, such as detecting whether there are animals approaching a campsite, whether there are animals attacking campers, etc., the solution of this embodiment can be applied in this type of application scenario. In training for the purpose of performing reconnaissance tasks, training task data is obtained, such as taking the protection of camping site A as the goal, detecting whether there are designated animals in area B with a radius of x1 centered on area A, detecting whether there are designated animals in area C with a radius of x2, and if so, closely monitoring the trajectory of such animals, detecting whether there are designated animals in area D with a radius of x3, and if so, outputting an alarm or killing such animals, and when detecting animals attacking campers in areas B, C, and D, the animals are killed, wherein the attack method is a marking bullet, and the animals in training can all be replaced by training dogs for simulation. After the system obtains the training task data, it performs deep semantic analysis on it and determines the training task. In order to send different types of drones to conduct site reconnaissance, the training task type is reconnaissance type. The task details include reconnaissance with area A as the center and whether a specified type of animal appears in area B with a radius of x1, reconnaissance with a radius of x2, whether a specified type of animal appears in area C, if so, closely monitor the trajectory of the animal, reconnaissance with a radius of x3, whether a specified type of animal appears in area D, if so, output an alarm or kill the animal. When an animal is detected attacking a camper in area B, C, or D, the animal is killed, and the attack method is a marking bullet. The animals in training can all be replaced by training dogs. Then the system determines the target task type that meets the mapping conditions based on the training task type, such as a reconnaissance task. At this time, the corresponding candidate calculation model is selected and determined based on the type of reconnaissance task. The model is used to calculate the reconnaissance data, the number and type of reconnaissance aircraft required, and the number of reconnaissance aircraft distributed in different areas (the reconnaissance aircraft is a drone with monitoring function). In order to make the candidate calculation model more accurately allocate the number of different types of reconnaissance aircraft, such as the number allocation and regional deployment of single reconnaissance aircraft, single attack aircraft, and reconnaissance + attack mixed aircraft, the system can adjust the parameters of the candidate calculation model according to the above-mentioned detailed information, such as adjusting the weights of the parameters corresponding to different types of aircraft, so that the adjusted target calculation model can provide more efficient and accurate calculations when executing training tasks in the scenario of this embodiment. After the model adjustment is completed, the training task for which the system has interpreted the specific information can be executed, and the target calculation model can be called to calculate the data calculation tasks involved in the middle, and finally determine whether the state after the allocation and deployment of resources and sites can support the smooth execution of the training task. If it can, each reconnaissance aircraft can be controlled for training based on the data after the simulation execution. If not, error information is obtained, such as which link is not effectively monitored or attacked, etc., and then the error feedback mechanism is executed based on the error information to regenerate the target calculation model in combination with the error feedback result.
[0093] like Figure 3 As shown, another embodiment of the present invention also provides a training task data processing device, including:
[0094] A response module, configured to respond to the obtained training task data and perform deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language;
[0095] A first determination module is used to determine a target task type that satisfies a mapping condition with the training task according to the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task;
[0096] A second determination module, used to determine a candidate computing model according to the target task type;
[0097] A first adjustment module, configured to adjust the candidate computing model according to the detail information to generate a target computing model;
[0098] A calling module is used to call the target computing model to try to execute the training task, complete the data calculation required for the task, and obtain the execution result;
[0099] The error feedback module is used to execute the error feedback mechanism when the execution result indicates that the task execution fails, so as to regenerate the target computing model in combination with the collected error information.
[0100] In some embodiments, the apparatus further comprises:
[0101] The second adjustment module is used to adjust the training task, training task type, and detailed information of the training task based on a preset output format to form structured data, and the structured data is in a data format that can be recognized by the computing model.
[0102] In some embodiments, determining the target task type that satisfies the mapping condition with the training task based on the training task type includes:
[0103] Pre-build multiple different predefined task types based on different task types;
[0104] Based on the training task type, a target task type corresponding to the training task is determined by matching among a plurality of predefined task types constructed.
[0105] In some embodiments, determining a candidate computing model based on the target task type includes:
[0106] Based on the target task type, a corresponding computing model is determined as a candidate computing model, and a model code of the candidate computing model is determined.
[0107] In some embodiments, adjusting the candidate computing model based on the detail information to generate a target computing model includes:
[0108] Adjusting parameters of the model code of the candidate computing model based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information, so as to generate the target computing model based on the adjusted model code; or
[0109] Determine the required toolkit based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information;
[0110] The toolkit is imported into the model code of the candidate computing model to generate the target computing model based on the adjusted model code.
[0111] In some embodiments, each of the predefined task types has a corresponding example code involved in performing the task of that type;
[0112] The device also includes:
[0113] The second adjustment module is used to adjust the sample code according to the detailed information, and configure the sample code to form a target code that matches the training task and can call the target computing model.
[0114] In some embodiments, calling the target computing model to try to execute the training task includes:
[0115] The target code is executed based on a coding parser, and the target computing model is called to perform computing tasks during the execution of the target code. The computing tasks are different for different training tasks.
[0116] In some embodiments, when the execution result indicates that the task execution fails, executing the error feedback mechanism includes:
[0117] When the execution result indicates that the task has failed, error information is collected;
[0118] The target code and / or target computing model is modified based on the error information to form a new target code and / or target computing model.
[0119] Another embodiment of the present invention also provides an electronic device, including:
[0120] at least one processor; and,
[0121] a memory communicatively connected to the at least one processor; wherein,
[0122] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to implement the training task data processing method as described in any one of the embodiments above.
[0123] Another embodiment of the present invention further provides a storage medium, which includes a stored program, wherein when the program is executed, a device including the storage medium is controlled to execute the automatic pipeline path selection method as described in any one of the above embodiments.
[0124] The embodiment of the present invention also provides a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions, and when the computer-executable instructions are executed, at least one processor executes the pipeline path automatic selection method such as in the embodiment described above. It should be understood that each solution in this embodiment has the corresponding technical effect in the above method embodiment, which will not be repeated here.
[0125] It should be noted that the computer storage medium of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program configured to be used by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, antenna, optical cable, RF, etc., or any suitable combination of the above.
[0126] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.
[0127] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.
Claims
1. A training task data processing method, characterized in that: include: In response to the obtained training task data, performing deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language; Determine a target task type that satisfies a mapping condition with the training task based on the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task; Determining a candidate computing model based on the target task type; Adjusting the candidate computing model based on the detailed information to generate a target computing model; Calling the target computing model to try to execute the training task, completing the data calculation required for the task, and obtaining the execution result; When the execution result indicates that the task execution has failed, an error feedback mechanism is executed to regenerate the target computing model in combination with the collected error information.
2. The training task data processing method according to claim 1, characterized in that: The method further comprises: The training tasks, training task types, and detailed information of the training tasks are adjusted based on a preset output format to form structured data, where the structured data is in a data format that can be recognized by the computing model.
3. The training task data processing method according to claim 1, characterized in that: The determining, based on the training task type, a target task type that satisfies a mapping condition with the training task includes: Pre-build multiple different predefined task types based on different task types; Based on the training task type, a target task type corresponding to the training task is determined by matching among a plurality of predefined task types constructed.
4. The training task data processing method according to claim 3, characterized in that: The determining of the candidate computing model based on the target task type includes: Based on the target task type, a corresponding computing model is determined as a candidate computing model, and a model code of the candidate computing model is determined.
5. The training task data processing method according to claim 4, characterized in that: The step of adjusting the candidate computing model based on the detail information to generate a target computing model includes: Adjusting parameters of the model code of the candidate computing model based on the training resources, training site resource deployment, training requirements, and training objectives in the detailed information, so as to generate the target computing model based on the adjusted model code; or Determine the required toolkit based on the training resources, resource deployment of the training site, training requirements, and training objectives in the detailed information; The toolkit is imported into the model code of the candidate computing model to generate the target computing model based on the adjusted model code.
6. The training task data processing method according to claim 1, characterized in that: Each of the predefined task types has a corresponding sample code involved in executing the task of this type; The method further comprises: The example code is adjusted based on the detailed information, and the example code is configured to form a target code that matches the training task and can call the target computing model.
7. The training task data processing method according to claim 6, characterized in that: The calling of the target computing model to try to execute the training task includes: The target code is executed based on the encoding parser, and the target computing model is called to perform computing tasks during the execution of the target code. The computing tasks are different for different training tasks.
8. The training task data processing method according to claim 6, characterized in that: When the execution result indicates that the task execution fails, executing the error feedback mechanism includes: When the execution result indicates that the task has failed, error information is collected; The target code and / or target computing model is modified based on the error information to form a new target code and / or target computing model.
9. A training task data processing device, characterized in that: include: A response module, configured to respond to the obtained training task data and perform deep semantic analysis on the training task data to determine the training task, the training task type and detailed information of the training task, wherein the training task data is data described in natural language; A first determination module is used to determine a target task type that satisfies a mapping condition with the training task according to the training task type, wherein the target task type is one of a plurality of predefined task types, and different predefined task types correspond to different computing models for executing the training task; A second determination module, used to determine a candidate computing model according to the target task type; A first adjustment module, configured to adjust the candidate computing model according to the detail information to generate a target computing model; A calling module is used to call the target computing model to try to execute the training task, complete the data calculation required for the task, and obtain the execution result; The error feedback module is used to execute the error feedback mechanism when the execution result indicates that the task execution fails, so as to regenerate the target computing model in combination with the collected error information.
10. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to implement the training task data processing method as described in any one of claims 1-8.
Citation Information
Cited By
Discrimination method, model training method and device, discrimination system and related product
CN121980356A
