Task processing method and device based on large model, equipment and medium

Abstract

The invention provides a task processing method and device based on a large model, equipment and a medium, and relates to the technical field of artificial intelligence, in particular to the technical fields of natural language processing, deep learning, large models and the like. The method comprises the steps that a plurality of target tools used for achieving a plurality of functions are determined from a plurality of candidate tool sets for a target task, and the target task comprises a plurality of sub-tasks with an execution sequence; and according to the execution sequence, processing the task information of the target task by using the plurality of target tools to obtain a training sample.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the technical fields of natural language processing, deep learning, large models, and more specifically to a task processing method and apparatus, device, and medium based on large models. Background Art

[0002] With the development of deep learning technology, large language models (LLMs) have emerged. Large language models refer to language models with a large number of parameters and pre-trained data obtained by training on large-scale text data using deep learning technology. Large language models can be used for various natural language processing tasks such as dialogue generation and text generation to obtain natural, fluent and accurate text output. Summary of the invention

[0003] The present invention provides a task processing method, device, equipment and medium based on a large model.

[0004] According to one aspect of the present disclosure, a large model-based sample generation method is provided, comprising: for a target task, determining multiple target tools for realizing multiple functions from multiple candidate tool sets, wherein the target task includes multiple subtasks with an execution order; and processing the task information of the target task using the multiple target tools in accordance with the execution order to obtain a training sample.

[0005] According to another aspect of the present disclosure, a large model training method is provided, comprising: inputting training samples into a large language model to be trained to obtain output information, wherein the training samples are generated using a sample generation method based on a large model; using the output information to train the large language model to be trained to obtain an initial large language model; using a target tool for realizing the effect evaluation function to evaluate the initial large language model to obtain a model evaluation result; and, when the model evaluation result indicates that the initial large language model has passed the evaluation, determining the initial large language model as a trained large language model.

[0006] According to another aspect of the present disclosure, a large model-based task processing method is provided, including: performing distillation training on a trained large language model to obtain a task processing model, wherein the trained large language model is trained using a large model training method; and sending model information of the task processing model to a client so that the client deploys the task processing model.

[0007] According to another aspect of the present disclosure, a sample generation device based on a large model is provided, including: a first determination module, for determining, in response to the above-mentioned target task, multiple target tools for realizing multiple functions from multiple candidate tool sets, wherein the above-mentioned target task includes multiple subtasks with an execution order; a first processing module, for processing the task information of the above-mentioned target task using the above-mentioned multiple target tools in accordance with the above-mentioned execution order to obtain a training sample.

[0008] According to another aspect of the present disclosure, a large model training device is provided, comprising: a first input module, used to input training samples into a large language model to be trained to obtain output information, wherein the training samples are generated by using a sample generation device based on a large model; a first training module, used to train the large language model to be trained using the output information to obtain an initial large language model; a first evaluation module, used to evaluate the initial large language model using a target tool for realizing the effect evaluation function to obtain a model evaluation result; and a second determination module, used to determine the initial large language model as a trained large language model when the model evaluation result indicates that the initial large language model has passed the evaluation.

[0009] According to another aspect of the present disclosure, a task processing device based on a large model is provided, including: a third training module, used to perform distillation training on a trained large language model to obtain a task processing model, wherein the trained large language model is trained using a large model training device; a sending module, used to send model information of the task processing model to a client, so that the client can deploy the task processing model.

[0010] According to another aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0011] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the above method are implemented.

[0012] According to another aspect of the present disclosure, a computer program product is provided, including a computer program or instructions, and the steps of the above method are implemented when the computer program or instructions are executed by a processor.

[0013] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0015] Figure 1 The system architecture of the sample generation, training and task processing methods based on a large model according to the embodiment of the present disclosure is schematically shown;

[0016] Figure 2 The flowchart of the sample generation method based on the large model according to the embodiment of the present disclosure is schematically shown;

[0017] Figure 3 An example schematic diagram schematically shows a process of obtaining task information according to an embodiment of the present disclosure;

[0018] Figure 4 An example schematic diagram schematically shows a target tool determination process according to an embodiment of the present disclosure;

[0019] Figure 5 An example schematic diagram of a process of obtaining training samples by processing task information using multiple target tools in an execution order according to an embodiment of the present disclosure is shown schematically;

[0020] Figure 6 A flowchart of a large model training method according to an embodiment of the present disclosure is schematically shown;

[0021] Figure 7 An example schematic diagram schematically shows a process of obtaining a trained large language model according to an embodiment of the present disclosure;

[0022] Figure 8 A flowchart of a task processing method based on a large model according to an embodiment of the present disclosure is schematically shown;

[0023] Fig. 9 A block diagram of a sample generation device based on a large model according to an embodiment of the present disclosure is schematically shown;

[0024] Fig.10 A block diagram of a large model training device according to an embodiment of the present disclosure is schematically shown;

[0025] Fig.11 A block diagram schematically shows a task processing device based on a large model according to an embodiment of the present disclosure and

[0026] Fig.12 The block diagram of an electronic device suitable for implementing a large-model-based sample generation, large-model training, and task processing method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0027] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0028] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0029] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0030] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0031] In the process of using the large language model, different business systems can obtain model information that meets different task requirements by connecting to the large language model, and deploy it to the local business system to realize task processing in the local business system.

[0032] However, since different business systems have different requirements for large language models, how to adaptively implement iterative training of large language models for different business systems while meeting the requirements of usage cost, model effect and iterative optimization efficiency is an urgent problem to be solved.

[0033] To this end, the embodiment of the present disclosure proposes a sample generation scheme based on a large model. For example, for a target task, multiple target tools for realizing multiple functions are determined from multiple candidate tool sets, wherein the target task includes multiple subtasks with an execution order; according to the execution order, the task information of the target task is processed using multiple target tools to obtain training samples.

[0034] According to the embodiments of the present disclosure, since multiple target tools for realizing multiple functions are determined from multiple candidate tool sets for target tasks, the selection of target tools can be dynamically adjusted according to the requirements of different target tasks, thereby improving the flexibility of task execution. By processing task information according to the execution order of multiple subtasks and the target tools for realizing corresponding functions to obtain training samples, the quality of training samples can be improved.

[0035] In the technical solution of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0036] In the technical solution of the present invention, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0037] Figure 1 The schematic diagram shows the system architecture of the sample generation, training and task processing methods based on the large model according to the embodiment of the present disclosure. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0038] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0039] The user may use at least one of the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0040] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0041] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0042] It should be noted that the sample generation based on the large model, the training of the large model, and the task processing method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the sample generation based on the large model, the training of the large model, and the task processing device provided in the embodiment of the present disclosure can generally be set in the server 105. The sample generation based on the large model, the training of the large model, and the task processing method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the sample generation based on the large model, the training of the large model, and the task processing device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0043] Alternatively, the large model-based sample generation, large model training, and task processing methods provided in the embodiments of the present disclosure may also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or may also be executed by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Accordingly, the large model-based sample generation, large model training, and task processing apparatus provided in the embodiments of the present disclosure may also be set in the first terminal device 101, the second terminal device 102, or the third terminal device 103, or may be set in other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103.

[0044] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0045] It should be noted that the sequence numbers of the operations in the following method are only used as representations of the operations for the purpose of description, and should not be regarded as representing the execution order of the operations. Unless explicitly stated, the method does not need to be executed completely in the order shown.

[0046] The above describes the system architecture of the sample generation method based on the large model provided by the present disclosure. Figure 2 For example, the task execution process of the present disclosure is further explained.

[0047] Figure 2 The flowchart of the sample generation method based on the large model according to the embodiment of the present disclosure is schematically shown.

[0048] like Figure 2 As shown, the task execution method 200 includes operations S210 to S220.

[0049] In operation S210 , for a target task, a plurality of target tools for implementing a plurality of functions are determined from a plurality of candidate tool sets, wherein the target task includes a plurality of subtasks having an execution order.

[0050] In operation S220, task information of the target task is processed using a plurality of target tools in an execution order to obtain a training sample.

[0051] The target task may include multiple subtasks. Subtasks may refer to smaller, specific task units decomposed from the target task. These subtasks have a specific execution order. The execution order refers to the need to execute these subtasks step by step in a certain order when completing the entire target task. For example, the target task includes subtask 1->subtask 2->subtask 3. In the process of executing the target task, subtask 1 needs to be executed first. After subtask 1 is executed, subtask 2 is executed. After subtask 2 is executed, subtask 3 is executed to complete the execution of the target task.

[0052] Multiple subtasks can be used to implement multiple functions. Multiple functions can be configured according to actual business needs and are not limited here. Multiple functions may include at least one of the following: prompt optimization function, question generation function, answer generation function, quality control function and effect evaluation function. For example, each subtask can be used to implement a different function, that is, subtask 1 is used to implement the prompt optimization function, subtask 2 is used to implement the question generation function, subtask 3 is used to implement the answer generation function, etc. Alternatively, at least two of the multiple subtasks can also be used to implement the same function, that is, subtask 1 and subtask 2 are both used to implement the prompt optimization function, and subtask 3 is used to implement the quality control function, etc.

[0053] The plurality of candidate tool sets may be pre-configured and can be used to implement a plurality of functions, i.e., the plurality of candidate tools in the same candidate tool set are used to implement the same function, and the plurality of candidate tools in different candidate tool sets are used to implement different functions. Each candidate tool set includes a set of candidate tools that can be selected for use. The candidate tools may refer to various software tools, hardware tools or methods that can be used to implement a specific function.

[0054] For a target task, multiple target tools for realizing multiple functions can be determined from multiple candidate tool sets. The method for determining the target tool can be configured according to actual business needs and is not limited here. For example, for each subtask, a candidate tool set corresponding to the function can be determined based on the function that needs to be realized by the subtask, and a target tool can be randomly selected from the multiple candidate tools included in the candidate tool set. Alternatively, for each candidate tool set, the tool with the best effect from the multiple candidate tools included in the candidate tool set can be pre-set, and the tool with the best effect corresponding to the function can be determined as the target tool based on the function that needs to be realized by each subtask.

[0055] After obtaining multiple target tools, the task information can be processed based on the execution order for each subtask using the target tool used to implement the function of the subtask to obtain a training sample. For example, in subtask 1, the task information can be processed using target tool 1 used to implement the function of subtask 1 to obtain first intermediate information. In subtask 2, the task information can be processed using target tool 2 used to implement the function of subtask 2 to obtain second intermediate information. In subtask 3, the first intermediate information and the second intermediate information can be processed using target tool 3 used to implement the function of subtask 3 to obtain a training sample. A training sample refers to a sample required for training a large language model.

[0056] Task information can be obtained by processing business data in a specific application domain. The method for obtaining business data can be configured according to actual business needs and is not limited here. For example, the business data can be directly obtained from the business system in a specific application domain using the Hypertext Transfer Protocol (HTTP). Alternatively, the business data in a specific application domain can be retrieved from a database by executing the Structured Query Language (SQL).

[0057] According to the embodiments of the present disclosure, since multiple target tools for realizing multiple functions are determined from multiple candidate tool sets for target tasks, the selection of target tools can be dynamically adjusted according to the requirements of different target tasks, thereby improving the flexibility of task execution. By processing task information according to the execution order of multiple subtasks and the target tools for realizing corresponding functions to obtain training samples, the quality of training samples can be improved.

[0058] The above describes the sample generation process based on the large model provided by the present disclosure. In the embodiment of the present disclosure, the task information can be obtained based on the initial business data of the application field included in the target task. Figure 3 The process of obtaining task information is further explained.

[0059] Figure 3 The following is a schematic diagram schematically showing an example of a task information obtaining process according to an embodiment of the present disclosure.

[0060] In one example, the target task 301 may be triggered in response to a timed task being triggered. A timed task may refer to a pre-set time interval or trigger condition, and a corresponding task or operation will be executed when the time interval or trigger condition meets a specific condition. For example, if the time interval is set to daily, weekly, or monthly, the timed task will be executed once a day, weekly, or monthly, and the corresponding target task 301 will also be triggered once a day, weekly, or monthly.

[0061] The setting method of the scheduled task and the target task can be configured according to actual business needs and is not limited here. For example, you can use the scheduled task management tool provided by the operating system or the scheduled task library of the programming language to set the time interval or trigger condition of the scheduled task, and combine the code of the target task with the scheduled task. In this way, when the set time interval or trigger condition is reached, the scheduled task is automatically triggered, thereby ensuring that the target task can be executed when the scheduled task is triggered.

[0062] In another example, target task 301 may be triggered by receiving a task creation instruction. A task creation instruction may refer to a clear instruction or event that instructs an operating system to create a new task or start a process for a specific task. A task creation instruction may be generated based on input from a user, other system, or automation tool.

[0063] The detection method of the task creation instruction can be configured according to actual business needs and is not limited here. For example, a trigger can be set through programming or automation tools to monitor specific events or instructions, and when it is determined that the task creation instruction is monitored, the target task is automatically triggered, thereby ensuring that the target task can be executed when the task creation instruction is received.

[0064] According to the embodiments of the present disclosure, by selecting trigger modes for different target tasks according to actual business needs and combining scheduled tasks with user instructions to automatically trigger target tasks, the response capability and task execution efficiency can be improved.

[0065] like Figure 3 As shown in 300 , the target task 301 may include initial business data 301_1 in the application domain.

[0066] Initial business data 301_1 may refer to business-related data collected in a specific application field. The application field may include any field in Natural Language Processing (NLP). For example, the application field may include at least one of the following: interview scenario, contract intelligent review scenario, sales scenario, quality inspection scenario, intelligent customer service, machine translation, intelligent recommendation, information retrieval, medical assistance, intelligent writing assistant, customer service and virtual assistant, etc. Initial business data 301_1 may include at least one of the following: user data, transaction data, operation data or marketing data, etc.

[0067] For the initial business data 301_1 in the target task 301, the initial business data 301_1 may be preprocessed to obtain processed data 304. The preprocessing may include at least one of the following: data cleaning, data verification, and data checking. In this case, the initial business data 301_1 may be subjected to data cleaning to obtain first intermediate business data 302, the first intermediate business data 302 may be subjected to data verification to obtain second intermediate business data 303, and the second intermediate business data 303 may be subjected to data verification to obtain processed data 304.

[0068] Data cleaning may include at least one of the following: duplicate value filtering, outlier filtering, missing value processing, and data formatting. Duplicate value filtering refers to eliminating duplicate records in the data to avoid data redundancy and improve data quality. Outlier value filtering refers to cleaning the data that contains erroneous, incomplete, or invalid information. Missing value processing refers to filling missing values ​​using statistical methods (such as mean, median, mode). Data formatting refers to using regular expressions or functions to handle formatting issues.

[0069] Data validation may include at least one of the following: text coherence analysis and verifying whether the data meets specific rules. Text coherence analysis refers to evaluating the fluency and coherence of the data. For example, a script or program can be used to automatically check the format and structure of the data. Verifying whether the data meets specific rules refers to using a data validation tool or library to validate the data based on specific rules. For example, the specific rule may be whether the data is within a preset range.

[0070] Data validation can include core field validation. Core field validation refers to the verification of key data fields to ensure that their format, scope, consistency, etc. meet the expected requirements. For example, regular expressions, conditional statements, or preset validation rules can be used to validate field content.

[0071] After obtaining the processed data 304, the processed data 304 may be encrypted to obtain the task information 305, so that all the data on the disk is ciphertext, thereby protecting data privacy. The specific encryption and decryption algorithm may be configured according to actual business needs and is not limited here. The encryption and decryption algorithm may include one of the following: Advanced Encryption Standard (AES) and symmetric encryption algorithm (Rivest-Shamir-Adleman, RSA).

[0072] For example, a key may be generated according to an encryption and decryption algorithm, and the processed data 304 may be encrypted using the key and based on the encryption algorithm to obtain task information 305. When data needs to be accessed or processed, the task information 305 may be processed using the key and the decryption algorithm to obtain the original task information.

[0073] According to the embodiments of the present disclosure, by performing pre-processing operations such as data cleaning, data verification and data checking on the initial business data, the quality of the data can be significantly improved. On this basis, by encrypting the processed data, sensitive information can be effectively protected. Based on this standardized data processing process, it is possible to quickly adapt to changes in the business environment and maintain the efficiency and accuracy of task information.

[0074] The above describes the process of obtaining the task information provided by the present disclosure. In the embodiment of the present disclosure, the target tool can be determined based on the tool identification, application field or historical usage information. Figure 4 Further explain the process of determining the target tool.

[0075] Figure 4 An example schematic diagram of a target tool determination process according to an embodiment of the present disclosure is schematically shown.

[0076] In one example, the plurality of functions may include at least one of the following: a prompt optimization function, a question generation function, an answer generation function, a quality control function, and an effect evaluation function.

[0077] The prompt optimization function is to help the model generate more coherent and accurate sentences or paragraphs by adjusting the input prompt information to obtain the prompt text. The question generation function is to generate appropriate question descriptions by processing the interactive information between the user and the model to obtain the question text. The answer generation function is to generate accurate answers corresponding to the question text by processing the interactive information between the user and the model to obtain the answer text. The quality control function is to ensure the quality, content accuracy and logical coherence of the generated text by limiting the generation length, introducing additional constraints, or using automatic review. The effect evaluation function is to evaluate the performance and results of the model on various evaluation indicators.

[0078] like Figure 4 As shown, in 400, taking multiple candidate tool sets including a candidate tool set 401 for implementing a prompt optimization function, a candidate tool set 402 for implementing a question generation function, a candidate tool set 403 for implementing an answer generation function, a candidate tool set 404 for implementing a quality control function, and a candidate tool set 405 for implementing an effect evaluation function as an example, the process of determining multiple target tools is explained.

[0079] It should be noted in advance that multiple candidate tools in the same candidate tool set are used to implement the same function, and multiple candidate tools in different candidate tool sets are used to implement different functions. Each candidate tool may refer to various software tools, hardware tools or methods.

[0080] In one example, for the candidate tool set 401, multiple candidate tools for implementing the same prompt optimization function include candidate tool 401_1, candidate tool 401_2, ..., candidate tool 401_M. M is a positive integer. Each candidate tool can have a candidate tool identifier, and the object can select to use a certain tool by specifying a specific identifier or mark, thereby determining the prompt optimization tool 406 from the multiple candidate tools in the candidate tool set 401 according to the tool identifier selected by the object in the target task and the candidate tool identifier of each candidate tool.

[0081] In another example, for candidate tool set 402, multiple candidate tools for implementing the same question generation function include candidate tool 402_1, candidate tool 402_2, ..., candidate tool 402_N. N is a positive integer. Each candidate tool can have an application field respectively, thus, a question generation tool 407 can be determined from multiple candidate tools in candidate tool set 402 according to the application field of the target task and the application field of each candidate tool respectively. For candidate tool set 403, multiple candidate tools for implementing the same answer generation function include candidate tool 403_1, candidate tool 403_2, ..., candidate tool 403_P. P is a positive integer. Thus, a question generation tool 408 can be determined from multiple candidate tools in candidate tool set 403 according to the application field of the target task and the application field of each candidate tool respectively.

[0082] In another example, for the candidate tool set 404, multiple candidate tools for implementing the same quality control function include candidate tool 404_1, candidate tool 404_2, ..., candidate tool 404_Q. Q is a positive integer. Each candidate tool may have historical usage information, which may refer to the records and data of each candidate tool being used in the past. Thus, the pros and cons of each candidate tool may be evaluated based on the historical usage information of each of the multiple candidate tools, so as to determine the quality control tool 408 from the multiple candidate tools in the candidate tool set 404. For the candidate tool set 405, multiple candidate tools for implementing the same effect evaluation function include candidate tool 405_1, candidate tool 405_2, ..., candidate tool 405_S. S is a positive integer. The pros and cons of each candidate tool may be evaluated based on the historical usage information of each of the multiple candidate tools, so as to determine the effect evaluation tool 410 from the multiple candidate tools in the candidate tool set 405.

[0083] It should be noted that the above-mentioned method of determining the target tool for realizing multiple functions in each candidate tool set can be configured according to actual business needs and is not limited here. For example, the prompt optimization tool 406 can also be determined based on the application field or historical usage information. Alternatively, the question generation tool 407 and the question generation tool 408 can also be determined based on the application field or historical usage information. Alternatively, the quality control tool 408 and the effect evaluation tool 410 can also be determined based on the tool identification or application field.

[0084] According to the embodiments of the present disclosure, by constructing a plurality of candidate tool sets that can be assembled and used in a general manner, a set of candidate tool sets can serve different business scenarios at the same time, and the standardization of tool capabilities and process mechanisms can be achieved, so that they can be quickly applied to different business scenarios. On this basis, by selecting target tools according to different modes such as tool identification, application field or historical usage information, the needs of specific business scenarios can be met, the time cost of tool selection is reduced, and the selection accuracy of target tools is improved, thereby helping to improve overall efficiency.

[0085] The above describes the process of determining the target tool provided by the present disclosure. Figure 5 The process of obtaining training samples by processing task information using multiple target tools in the execution order is further described.

[0086] Figure 5 The diagram schematically shows an example of a process of obtaining training samples by processing task information using multiple target tools in an execution order according to an embodiment of the present disclosure.

[0087] like Figure 5 As shown, in 500, task information 501 may include prompt information 501_1 of the large language model and interaction information 501_2 obtained by the object interacting with the large language model. Prompt information 501_1 refers to information used to guide the large language model to generate output answers based on input questions. Interaction information 501_2 refers to content generated during the dialogue or communication between the user and the large language model, including input questions provided by the user, output answers generated by the model, and feedback information from the user on the output answers.

[0088] In one example, the prompt optimization tool 502 for implementing the prompt optimization function can be used to process the prompt information 501_1 to obtain the prompt text 503. The prompt optimization tool 502 can include at least one of the following: a natural language processing tool and a text generation model. For example, the prompt optimization tool 502 can be used to perform preprocessing and other operations on the prompt information 501_1 to obtain intermediate prompt information; extract key features from the intermediate prompt information to obtain prompt features; and process the prompt features to obtain the prompt text 503.

[0089] In another example, the interactive information 501_2 may be processed by a question generation tool 504 for implementing a question generation function to obtain a question text 505. The interactive information 501_2 may be processed by an answer generation tool 506 for implementing an answer generation function to obtain an answer text 507. The question generation tool 504 may include at least one of the following: a natural language processing tool and a text generation model. For example, the interactive information 501_2 may be preprocessed by the question generation tool 504 to obtain first intermediate interactive information; key features may be extracted from the first intermediate interactive information to obtain first interactive features; the first interactive features may be processed to obtain the question text 505. Alternatively, the interactive information 501_2 may be preprocessed by the answer generation tool 506 to obtain second intermediate interactive information; key features may be extracted from the second intermediate interactive information to obtain second interactive features; the second interactive features may be processed to obtain an answer text 507.

[0090] On this basis, the target tool for implementing the quality control function can be used to evaluate the question text 505 and the answer text 507 to obtain a quality evaluation result. The target tool for implementing the quality control function can include at least one of the following: a text analysis model, a machine learning algorithm, a data mining technology, and related software or platforms. For example, the features of the question text 505 and the answer text 507 can be extracted respectively to obtain question features and answer features; the correlation between the question features and the answer features can be determined to obtain a quality evaluation result.

[0091] After obtaining the quality assessment result, it can be determined whether the quality assessment result meets the predetermined condition. The predetermined condition can be configured according to actual business needs and is not limited here. For example, the predetermined condition can be that the correlation between the question text 505 and the answer text 507 is greater than or equal to a predetermined correlation threshold. In response to the quality assessment result indicating that the question text 505 and the answer text 507 meet the predetermined condition, a training sample 508 can be generated based on the prompt text 503, the question text 505 and the answer text 507.

[0092] According to the embodiments of the present disclosure, by using target tools for realizing different functions to carry out prompt optimization, question generation and answer generation in a targeted manner, the accuracy of prompt text, question text and answer text is ensured. On this basis, by using quality control tools to evaluate question text and answer text, and generating training samples based on the obtained quality evaluation results, the quality of training samples is improved, and a reliable data basis can be provided for subsequent model training. In addition, since training samples can be automatically generated, there is no need to invest a large amount of manual annotation of training data, which not only saves labor costs, but also speeds up data collection and expands the amount of data.

[0093] The above describes the process of obtaining the training samples provided by the present disclosure. In the embodiment of the present disclosure, after obtaining the training samples, the training samples can be used to train a trained large language model. Figure 6 Further explain the training process of the large model.

[0094] Figure 6 A flowchart of a large model training method according to an embodiment of the present disclosure is schematically shown.

[0095] like Figure 6 As shown, the large model training method 600 includes operations S610 to S640.

[0096] In operation S610, a training sample is input into a large language model to be trained to obtain output information, wherein the training sample is generated by using a sample generation method based on a large model.

[0097] In operation S620, the large language model to be trained is trained using the output information to obtain an initial large language model.

[0098] In operation S630, the initial large language model is evaluated using a target tool for implementing an effect evaluation function to obtain a model evaluation result.

[0099] In operation S640 , if the model evaluation result indicates that the initial large language model passes the evaluation, the initial large language model is determined as a trained large language model.

[0100] After obtaining the training samples, the large language model can be trained using the training samples to obtain a trained large language model, which refers to a large language model that can meet the requirements of the target task. The model evaluation result can characterize the model performance obtained by evaluating the training samples. The model evaluation result may include at least one of the following: accuracy, precision, recall, F1 score, AUC (area under the curve), etc. The training method may include at least one of the following: pruning, quantization, distillation, compression, and depth-separable convolution, etc.

[0101] According to the embodiments of the present disclosure, after obtaining training samples, the model effect can be quickly improved by using high-quality training samples for training models and using target tools for implementing effect evaluation functions to evaluate and iteratively train the trained models. Since the full process of training sample generation, model training, effect evaluation and deployment is realized in an automated closed loop, it is possible to connect to the business at a low cost and improve the efficiency of task execution.

[0102] Figure 7 An example schematic diagram of a process for obtaining a trained large language model according to an embodiment of the present disclosure is schematically shown.

[0103] like Figure 7 As shown, in 700, after obtaining training samples 701, the candidate large language model can be trained using the training samples 701 to obtain an initial large language model 702. The initial large language model 702 is evaluated using an effect evaluation tool 704 for implementing an effect evaluation function to obtain a model evaluation result 704.

[0104] After obtaining the model evaluation result 704, operation S710 may be performed. In operation S710, it is determined whether the model evaluation result 704 indicates that the initial large language model 702 passes the evaluation.

[0105] If so, the initial large language model 702 can be trained to obtain a task processing model with fewer model parameters and model information 705 of the task processing model. When the model evaluation result 704 indicates that the initial large language model 702 has passed the evaluation, it means that the effect of the initial large language model 702 has shown a positive improvement, and thus the initial large language model 702 can be determined as the trained large language model 705.

[0106] If not, multiple update tools 706 can be re-determined from multiple candidate tool sets. Based on the execution order, multiple update tools 706 are used to process the task information to obtain updated training samples 707. The initial large language model 702 is trained using the updated training samples 707 to obtain an updated large language model 708. The updated large language model 708 is evaluated using an updated effect evaluation tool 709 for implementing an effect evaluation function to obtain an updated model evaluation result 710. The updated model evaluation result 710 can characterize the model performance evaluated using the updated training samples 707. The updated model evaluation result 710 can include at least one of the following: accuracy, precision, recall, F1 score, AUC (area under the curve), etc.

[0107] After obtaining the updated model evaluation result 710, operation S720 may be performed. In operation S720, it is determined whether the updated model evaluation result 710 indicates that the updated large language model 708 has passed the evaluation.

[0108] If so, the updated large language model 708 can be trained to obtain a task processing model with fewer model parameters and model information 705 of the task processing model. The training method may include at least one of the following: pruning, quantization, distillation, compression, and depthwise separable convolution. When the updated model evaluation result 710 indicates that the updated large language model 708 has passed the evaluation, it means that the effect of the updated large language model 708 is positively improved, and thus the initial large language model 702 can be determined as the trained large language model 705.

[0109] If not, the above operations may be repeatedly performed until the updated model evaluation result 710 indicates that the updated large language model 708 has passed the evaluation.

[0110] Figure 8 The flowchart of the task processing method based on a large model according to an embodiment of the present disclosure is schematically shown.

[0111] like Figure 8 As shown, the large model-based task processing method 800 includes operations S810 to S820.

[0112] In operation S810, distillation training is performed on the trained large language model to obtain a task processing model, wherein the trained large language model is trained using a large model training method.

[0113] In operation S820 , model information of the task processing model is sent to the client so that the client deploys the task processing model.

[0114] Since the trained large language model has more model parameters, a task processing model can be generated based on the trained large language model. The task processing model refers to a small model with fewer model parameters obtained by training the trained large language model. The method of generating the task processing model based on the large language model can be configured according to actual business needs and is not limited here. For example, the method of generating the task processing model can include at least one of the following: pruning, quantization, distillation, compression, and depth-separable convolution.

[0115] After obtaining the task processing model, the model information of the task processing model can be sent to the client's business system so that the business system can use the model information to deploy the task processing model locally to achieve task processing in the application field. The model information of the task processing model may include at least one of the following: architecture or type, number of parameters, number of layers and neuron structure, hyperparameter settings, performance indicators, data sets and data preprocessing methods used during training, optimizer selection and optimization algorithm, number of training iterations or duration, performance results on training sets, validation sets and test sets, etc.

[0116] According to the embodiments of the present disclosure, by adopting the model distillation solution, the task processing model is trained for business use based on the large model obtained by training with training samples, which effectively realizes the compression and optimization of the model, reduces the computing overhead, thereby reducing the deployment cost and improving the task processing efficiency. It should be noted that in the embodiments of the present invention, some existing solutions in the industry such as certain software, components, and models may be mentioned, which should be regarded as exemplary, and their purpose is only to illustrate the feasibility of the implementation of the technical solution of the present invention, but it does not mean that the present invention has or must use the solution.

[0117] The above are only exemplary embodiments, but are not limited thereto and may also include other task execution methods known in the art, as long as the selection of target tools can be dynamically adjusted according to the requirements of different target tasks to ensure task execution efficiency and the model effect of the task processing model.

[0118] Fig. 9 A block diagram of a sample generation device based on a large model according to an embodiment of the present disclosure is schematically shown.

[0119] like Fig. 9 As shown, the large model-based sample generation device 900 may include a first determination module 910 and a first processing module 920 .

[0120] The first determination module 910 is used to determine a plurality of target tools for implementing a plurality of functions from a plurality of candidate tool sets in response to a target task, wherein the target task includes a plurality of subtasks having an execution order.

[0121] The first processing module 920 is used to process the task information of the target task using multiple target tools in the execution order to obtain training samples.

[0122] According to an embodiment of the present disclosure, each candidate tool set includes a plurality of candidate tools for implementing the same function.

[0123] According to an embodiment of the present disclosure, the first determining module 910 includes one of the following: a first determining unit, a second determining unit, and a third determining unit.

[0124] The first determination unit is used to determine a target tool from a plurality of candidate tools according to a tool identifier selected by an object in a target task for each candidate tool set.

[0125] The second determining unit is used to determine a target tool from a plurality of candidate tools according to an application field of the target task.

[0126] The third determining unit is used to determine a target tool from among the multiple candidate tools according to the historical usage information of each of the multiple candidate tools.

[0127] According to an embodiment of the present disclosure, the task information includes prompt information of the large language model and interaction information obtained by the object interacting with the large language model.

[0128] According to an embodiment of the present disclosure, the first processing module 920 includes a first processing unit, a second processing unit, a third processing unit, an evaluation unit, and a generation unit.

[0129] The first processing unit is used to process the prompt information by using a target tool for realizing the prompt optimization function to obtain a prompt text.

[0130] The second processing unit is used to process the interactive information using a target tool for realizing the question generation function to obtain the question text.

[0131] The third processing unit is used to process the interactive information using a target tool for realizing an answer generation function to obtain an answer text.

[0132] The evaluation unit is used to evaluate the question text and the answer text by using a target tool for realizing the quality control function to obtain a quality evaluation result.

[0133] The generating unit is used to generate training samples based on the prompt text, the question text and the answer text in response to the quality assessment result indicating that the question text and the answer text meet a predetermined condition. According to an embodiment of the present disclosure, the target task includes initial business data of the application domain.

[0134] According to an embodiment of the present disclosure, the task information is obtained by: preprocessing the initial business data to obtain processed data, wherein the preprocessing includes at least one of the following: data cleaning, data verification and data checking. The processed data is encrypted to obtain the task information.

[0135] According to an embodiment of the present disclosure, the target task is triggered by one of the following ways: a scheduled task is triggered; or a task creation instruction is received.

[0136] According to an embodiment of the present disclosure, the plurality of functions include at least one of the following: a prompt optimization function, a question generation function, an answer generation function, a quality control function, and an effect evaluation function.

[0137] Fig.10 A block diagram of a large model training device according to an embodiment of the present disclosure is schematically shown.

[0138] like Fig.10 As shown, the large model training device 1000 may include

[0139] The first input module 1010 is used to input training samples into the large language model to be trained to obtain output information, wherein the training samples are generated by using a sample generation device based on the large model.

[0140] The first training module 1020 is used to train the large language model to be trained using the output information to obtain an initial large language model.

[0141] The first evaluation module 1030 is used to evaluate the initial large language model using a target tool for implementing an effect evaluation function to obtain a model evaluation result.

[0142] The second determination module 1040 is configured to perform distillation training on the initial large language model to obtain model information of the task processing model when the model evaluation result indicates that the initial large language model has passed the evaluation.

[0143] According to an embodiment of the present disclosure, the large model training device 1000 may further include a second training module and a second evaluation module.

[0144] The second training module is used to train the initial large language model using the updated training samples to obtain an updated large language model, wherein the updated training samples are regenerated using a sample generation device based on the large model. The second evaluation module is used to evaluate the updated large language model using an update tool for implementing an effect evaluation function to obtain an updated model evaluation result.

[0145] Fig.11 A block diagram of a task processing device based on a large model according to an embodiment of the present disclosure is schematically shown.

[0146] like Fig.11 As shown, the large model-based task processing device 1100 may include a third training module 1110 and a sending module 1120 .

[0147] The third training module 1110 is used to perform distillation training on the trained large language model to obtain a task processing model, wherein the trained large language model is trained using a large model training device.

[0148] The sending module 1120 is used to send the model information of the task processing model to the client so that the client can deploy the task processing model.

[0149] Fig.12 A block diagram of an electronic device suitable for implementing sample generation based on a large model, training of a large model, and a task processing method according to an embodiment of the present disclosure is schematically shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0150] like Fig.12As shown, the device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 into a random access memory (RAM) 1203. In the RAM 1203, various programs and data required for the operation of the device 1200 can also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0151] A number of components in the device 1200 are connected to the I / O interface 1205, including: an input unit 1206, such as a keyboard, a mouse, etc.; an output unit 1207, such as various types of displays, speakers, etc.; a storage unit 1208, such as a disk, an optical disk, etc.; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1209 allows the device 1200 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0152] The computing unit 1201 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1201 performs the various methods and processes described above, such as the task execution method. For example, in some embodiments, the task execution method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1200 via the ROM 1202 and / or the communication unit 1209. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the task execution method described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to execute the task execution method in any other appropriate manner (e.g., by means of firmware).

[0153] Various implementations 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 chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0155] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer 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 can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 acoustic input, voice input, or tactile input).

[0157] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0158] A computer system may include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises through computer programs running on respective computers and having a client-server relationship to each other. The server may be a cloud server, a server in a distributed system, or a server combined with a blockchain.

[0159] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0160] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A sample generation method based on a large model, comprising: For a target task, determining a plurality of target tools for implementing a plurality of functions from a plurality of candidate tool sets, wherein the target task includes a plurality of subtasks having an execution order; and According to the execution order, the task information of the target task is processed using the multiple target tools to obtain training samples.

2. The method according to claim 1, wherein: Each of the candidate tool sets includes a plurality of candidate tools for implementing the same function; Determining, for the target task, from a plurality of candidate tool sets, a plurality of target tools for implementing a plurality of functions comprises one of the following: For each of the candidate tool sets, determining the target tool from among the plurality of candidate tools according to a tool identifier selected by the subject in the target task; or Determining the target tool from among the multiple candidate tools according to the application field of the target task; or The target tool is determined from among the plurality of candidate tools according to the historical usage information of each of the plurality of candidate tools.

3. The method according to claim 1, wherein: The task information includes prompt information of the large language model and interaction information obtained by the object interacting with the large language model; The processing of the task information using the multiple target tools in the execution order to obtain the training samples includes: Using a target tool for realizing the prompt optimization function, the prompt information is processed to obtain a prompt text; Using a target tool for realizing the question generation function, the interactive information is processed to obtain a question text; Using a target tool for realizing the answer generation function, the interactive information is processed to obtain an answer text; Using a target tool for implementing the quality control function, the question text and the answer text are evaluated to obtain a quality evaluation result; and In response to the quality assessment result indicating that the question text and the answer text meet a predetermined condition, the training sample is generated based on the prompt text, the question text and the answer text.

4. The method according to claim 1, wherein: The target task includes initial business data of the application field; The task information is obtained in the following way: Preprocessing the initial business data to obtain processed data, wherein the preprocessing includes at least one of the following: data cleaning, data verification and data checking; and The processed data is encrypted to obtain the task information.

5. The method according to claim 1, wherein: The target task is triggered by one of the following methods: The scheduled task is triggered; or Received a task creation instruction.

6. The method according to any one of claims 1 to 5, wherein: The multiple functions include at least one of the following: a prompt optimization function, a question generation function, an answer generation function, a quality control function and an effect evaluation function.

7. A large model training method, comprising: Inputting a training sample into a large language model to be trained to obtain output information, wherein the training sample is generated using the method of any one of claims 1 to 6; Using the output information, training the large language model to be trained to obtain an initial large language model; Using a target tool for implementing the effect evaluation function, the initial large language model is evaluated to obtain a model evaluation result; and When the model evaluation result indicates that the initial large language model passes the evaluation, the initial large language model is determined as a trained large language model.

8. The method according to claim 7, further comprising: If the model evaluation result indicates that the initial large language model has not passed the evaluation, repeatedly perform the following operations until the updated model evaluation result indicates that the updated large language model has passed the evaluation: Using updated training samples, training the initial large language model to obtain an updated large language model, wherein the updated training samples are regenerated using the method according to any one of claims 1 to 6; The updated large language model is evaluated using an update tool for implementing an effect evaluation function to obtain an updated model evaluation result.

9. A task processing method based on a large model, comprising: Performing distillation training on the trained large language model to obtain a task processing model, wherein the trained large language model is trained using the method described in any one of claims 7 and 8; The model information of the task processing model is sent to the client so that the client can deploy the task processing model.

10. A sample generation device based on a large model, comprising: a first determining module, configured to determine, in response to the target task, a plurality of target tools for implementing a plurality of functions from a plurality of candidate tool sets, wherein the target task includes a plurality of subtasks having an execution order; and The first processing module is used to process the task information of the target task using the multiple target tools in accordance with the execution order to obtain a training sample.

11. The device according to claim 10, wherein: Each of the candidate tool sets includes a plurality of candidate tools for implementing the same function; The first determining module includes one of the following: A first determining unit is configured to determine, for each of the candidate tool sets, the target tool from among the plurality of candidate tools according to a tool identifier selected by the subject in the target task; or A second determining unit, configured to determine the target tool from among the plurality of candidate tools according to an application field of the target task; or The third determining unit is used to determine the target tool from the multiple candidate tools according to the historical usage information of each of the multiple candidate tools.

12. The device according to claim 10, wherein: The task information includes prompt information of the large language model and interaction information obtained by the object interacting with the large language model; The first processing module comprises: A first processing unit, configured to process the prompt information using a target tool for realizing the prompt optimization function to obtain a prompt text; A second processing unit, configured to process the interactive information using a target tool for realizing the question generation function to obtain a question text; A third processing unit, configured to process the interaction information using a target tool for realizing the answer generation function to obtain an answer text; an evaluation unit, configured to evaluate the question text and the answer text using a target tool for implementing the quality control function to obtain a quality evaluation result; and A generating unit is used for generating the training sample based on the prompt text, the question text and the answer text in response to the quality assessment result indicating that the question text and the answer text meet a predetermined condition.

13. The device according to claim 10, wherein: The target task includes initial business data of the application field; The task information is obtained in the following way: Preprocessing the initial business data to obtain processed data, wherein the preprocessing includes at least one of the following: data cleaning, data verification and data checking; and The processed data is encrypted to obtain the task information.

14. The device according to claim 10, wherein: The target task is triggered by one of the following methods: The scheduled task is triggered; or Received a task creation instruction.

15. The device according to any one of claims 10 to 14, wherein: The multiple functions include at least one of the following: a prompt optimization function, a question generation function, an answer generation function, a quality control function and an effect evaluation function.

16. A large model training device, comprising: A first input module, used for inputting a training sample into a large language model to be trained to obtain output information, wherein the training sample is generated by using the apparatus according to any one of claims 10 to 15; A first training module is used to train the large language model to be trained using the output information to obtain an initial large language model; A first evaluation module is used to evaluate the initial large language model using a target tool for implementing the effect evaluation function to obtain a model evaluation result; and The second determination module is configured to determine the initial large language model as a trained large language model if the model evaluation result indicates that the initial large language model has passed the evaluation.

17. The apparatus according to claim 16, further comprising: A second training module, configured to train the initial large language model using updated training samples to obtain an updated large language model, wherein the updated training samples are regenerated using the apparatus according to any one of claims 10 to 15; The second evaluation module is used to evaluate the updated large language model using an update tool for implementing an effect evaluation function to obtain an evaluation result of the updated model.

18. A task processing device based on a large model, comprising: A third training module is used to perform distillation training on the trained large language model to obtain a task processing model, wherein the trained large language model is trained using the device described in any one of claims 16 and 17; The sending module is used to send the model information of the task processing model to the client so that the client can deploy the task processing model.

19. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 9.

20. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

21. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images