Text processing method and device based on large model, equipment, medium, program product and intelligent agent

Abstract

The invention provides a text processing method and device based on a large model, equipment, a medium, a program product and an intelligent agent, and relates to the technical field of artificial intelligence, in particular to the technical field of deep learning, large models and generative models. The text processing method based on the large model comprises the steps that in response to a data evaluation request, the large model is used for processing an initial text set and cue words determined based on a preset tag system, tags related to the initial text set are obtained, the preset tag system comprises multiple preset tags and attribute relations among the multiple preset tags, and the attribute relations of the multiple preset tags are obtained; the preset label system is determined by processing a historical text set based on a large model; and performing attribute detection on the initial text set based on the label to obtain attribute evaluation information, the attribute evaluation information representing an expected training effect of executing the specified large model training task based on the initial text set.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the field of deep learning, large models and generative model technology, and can be applied to scenarios such as model training and model fine-tuning. Specifically, it relates to a text processing method, device, equipment, medium, program product and intelligent agent based on a large model. Background Art

[0002] With the development of artificial intelligence technology, data sets have become the basis for models to learn effectively. Models can be trained or fine-tuned based on data sets to improve their performance or make them suitable for specific scenarios. Summary of the invention

[0003] The present disclosure provides a text processing method, apparatus, device, medium, program product and intelligent agent based on a large model.

[0004] According to one aspect of the present disclosure, a text processing method is provided, comprising: in response to a data evaluation request, using a large model to process an initial text set and prompt words determined based on a preset label system to obtain labels related to the initial text set, the preset label system comprising multiple preset labels and attribute relationships between the multiple preset labels, the preset label system being determined based on the large model processing of a historical text set; performing attribute detection on the initial text set based on the labels to obtain attribute evaluation information, the attribute evaluation information representing an expected training effect of executing a specified large model training task based on the initial text set.

[0005] According to another aspect of the present disclosure, an artificial intelligence agent is provided, which is configured to execute the method provided according to the embodiment of the present disclosure.

[0006] According to another aspect of the present disclosure, a text processing device is provided, including: a determination module, for responding to a data evaluation request, using a large model to process an initial text set and prompt words determined based on a preset label system, to obtain labels related to the initial text set, the preset label system including multiple preset labels and attribute relationships between the multiple preset labels, the preset label system being determined based on the large model processing of a historical text set; a detection module, for performing attribute detection on the initial text set based on the labels to obtain attribute evaluation information, the attribute evaluation information representing an expected training effect of executing a specified large model training task based on the initial text set.

[0007] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.

[0008] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method as described above.

[0009] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the method described above when executed by a processor.

[0010] 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

[0011] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0012] Figure 1 An exemplary system architecture to which a large model-based text processing method and apparatus can be applied according to an embodiment of the present disclosure is schematically shown;

[0013] Figure 2 The flowchart of the text processing method based on the large model according to the embodiment of the present disclosure is schematically shown;

[0014] Figure 3 An initial text set and tags related to the initial text set according to an embodiment of the present disclosure are exemplarily shown;

[0015] Figure 4 A schematic diagram of determining a target training strategy according to an embodiment of the present disclosure is exemplified;

[0016] Figure 5 A large model-based data processing system according to an embodiment of the present disclosure is schematically shown;

[0017] Figure 6 A block diagram of a text processing device based on a large model according to an embodiment of the present disclosure is schematically shown;

[0018] Figure 7 The structure diagram of the artificial intelligence agent according to the embodiment of the present disclosure is schematically shown.

[0019] Figure 8 A block diagram of an electronic device suitable for implementing a large model-based text processing method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0020] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0021] In the technical solution disclosed in the present invention, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations, necessary confidentiality measures are taken, and do not violate public order and good morals.

[0022] With the development of artificial intelligence, text sets can be labeled, and labeled text sets can be used to perform specified model training tasks to improve model performance or make the model suitable for specific scenarios. However, the inventors found that there are still problems with the lack of comprehensiveness and systematicness of labels, and the evaluation method for labeled text sets is relatively simple, which makes it difficult to meet more refined training needs.

[0023] The disclosed embodiments provide a text processing method, device, equipment, medium, program product and intelligent agent based on a big model. The text processing method based on a big model includes: in response to a data evaluation request, using a big model to process an initial text set and prompt words determined based on a preset label system to obtain labels related to the initial text set, the preset label system includes multiple preset labels and attribute relationships between the multiple preset labels, and the preset label system is determined based on the big model processing of a historical text set; performing attribute detection on the initial text set based on the labels to obtain attribute evaluation information, the attribute evaluation information representing the expected training effect of executing a specified big model training task based on the initial text set.

[0024] According to the embodiments of the present disclosure, by processing the historical text set through a large model to obtain a preset label system, it is possible to use the large model to fully mine the preset label content and preset label attribute information of the historical text set, and build a more accurate, comprehensive and multi-dimensional preset label system, which provides a basis for the subsequent determination of labels related to the initial text set based on the large model and attribute detection of the initial text set. By using the large model to process the initial text set and the prompt words determined based on the preset label system, the large model can be guided to output labels related to the initial text set, and the use of prompt words to control the large model can be based on the understanding of the initial text set. Further mine the semantic attributes of the initial text set, more accurately and comprehensively obtain the labels related to the initial text set, thereby improving the annotation precision and accuracy of the initial text set. By detecting the attribute evaluation information of the initial text set based on the label, it is possible to accurately evaluate the training effect of the initial text set to perform the specified training task, so as to improve the training effect of the subsequent use of the initial text set to perform the specified model training task.

[0025] Figure 1 An exemplary system architecture to which the content processing method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown.

[0026] It should be noted that Figure 1 The examples shown are only examples of system architectures 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 do not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. For example, in another embodiment, an exemplary system architecture to which the large model-based text processing method and apparatus can be applied may include a terminal device, but the terminal device may implement the large model-based text processing method and apparatus provided by the embodiments of the present disclosure without interacting with the server.

[0027] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0028] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (only as examples).

[0029] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0030] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for the content browsed by the user using the terminal devices 101, 102, and 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 the user request) to the terminal device.

[0031] The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server can also be a server of a distributed system, or a server combined with blockchain.

[0032] It should be noted that the text processing method based on the big model provided by the text processing device in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the text processing device based on the big model provided by the embodiment of the present disclosure can generally be set in the server 105. The text processing method based on the big model provided by 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 terminal devices 101, 102, 103 and / or the server 105. Accordingly, the text processing device based on the big model provided by 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 terminal devices 101, 102, 103 and / or the server 105.

[0033] 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.

[0034] Figure 2 The flowchart of the text processing method based on the large model according to the embodiment of the present disclosure is schematically shown.

[0035] like Figure 2 As shown, the large model-based text processing method 200 includes operations S210 to S220.

[0036] In operation S210, in response to a data evaluation request, the initial text set and prompt words determined based on a preset label system are processed using a large model to obtain labels related to the initial text set. The preset label system includes multiple preset labels and attribute relationships between the multiple preset labels. The preset label system is determined based on the large model processing the historical text set.

[0037] In operation S220, attribute detection is performed on the initial text set based on the label to obtain attribute evaluation information, where the attribute evaluation information represents the expected training effect of executing the specified large model training task based on the initial text set.

[0038] According to an embodiment of the present disclosure, the initial text set can be used to perform a training task for training a large model, and the training task can be, for example, a text classification task, a machine translation task, a sentiment analysis task, a text generation task, a question-answering task, etc. Exemplarily, the type of the initial text set can include, for example, news, articles, product information, social media comments, encyclopedia data, etc., and the form of the initial text set can include, for example, structured text (such as tabular data), unstructured text (such as articles, question-answering dialogues), semi-structured text (such as XML documents), etc.

[0039] According to an embodiment of the present disclosure, the preset tag system may include multiple preset tags and attribute relationships between the multiple preset tags. For example, the preset tags may be used to describe the length, field, scene, format, source, time, emotional tendency, style, source reliability, language, etc. of the text.

[0040] According to an embodiment of the present disclosure, the prompt word determined based on the preset tag system can guide the large model to determine the tag related to the initial text set from the preset tags of the preset tag system. For example, the prompt word can be input into the large model, and the large model can match the initial text set with the preset tag system and output the tag related to the initial text set.

[0041] According to the embodiments of the present disclosure, the large model may be a model having a large-scale parameter structure, and the parameter magnitude is generally in the tens of millions or hundreds of millions, and may reach the billions or tens of billions. In terms of the network structure of the large model, for example, a network structure such as UFO (Unified Feature Optimization) may be adopted. By processing the initial text set through the large model, the relatively powerful feature understanding and expression capabilities of the large model can be brought into play, and the initial text set with a large data scale can be efficiently processed, thereby improving the accuracy of the labels related to the initial text set and the efficiency of label determination.

[0042] According to the embodiments of the present disclosure, a prompt word can be understood as a tag sequence composed of text, symbols and other tags input to the large model in the large model, and the prompt word can guide the model to generate a specific output or perform a specific task. As an example, in natural language processing (NLP), a prompt word can be a question, the beginning of a sentence, an instruction, or any text that can stimulate the model to produce an expected response.

[0043] As an example, the prompt words determined based on the preset tag system can indicate the preset tag system (such as the identification and matching method of the preset tag system), the restriction conditions of the tags (such as the number, format, range, granularity, exclusions of tags), etc., thereby effectively guiding the large model to automatically mine tags related to the initial text set according to the preset tag system.

[0044] According to an embodiment of the present disclosure, the tag associated with the initial text set may include one or more tags, and the embodiment of the present disclosure does not limit the number of tags.

[0045] According to an embodiment of the present disclosure, attribute detection can be performed on the initial text set based on tags related to the initial text set to obtain attribute evaluation information. The attribute evaluation information can be used to characterize the expected training effect of performing a specified large model training task based on the initial text set.

[0046] According to the embodiments of the present disclosure, the specified large model training task can be understood as a training task for fine-tuning the large model for a specified application scenario or specified requirement. By executing the specified training task, the large model can more accurately serve the application requirements to be executed, thereby improving the effectiveness and reliability of the large model in the specified application scenario or task.

[0047] For example, the specified big model training tasks may include text prediction tasks for the big model in the fields of medical, legal, and financial fields. Or the specified big model training tasks may include text prediction tasks for the big model in specified scenarios such as customer service conversations, content creation, and educational counseling. Or the specified big model training tasks may also represent training tasks for specified text processing functions such as machine translation and sentiment analysis.

[0048] According to an embodiment of the present disclosure, the attribute evaluation information can characterize the complexity, diversity, balance, scale, timeliness and other attributes of the initial text set related to the training task of executing the large model. Complexity can characterize the convergence difficulty, training time, prediction accuracy and other indicators of the task of fine-tuning the large model in the process of training the large model based on the initial text set and labels. Attribute detection of the initial text set based on labels can include evaluating the complexity information of the initial text set based on the attribute information of the label such as the semantic complexity attribute, technical field attribute, and application scenario attribute of the label.

[0049] According to the embodiments of the present disclosure, diversity can characterize the scenarios, fields, topics, styles, etc. covered by the initial text set. A text set with higher diversity helps the model learn richer features and improve generalization capabilities. A text set lacking diversity may cause the model to perform poorly when faced with unseen data.

[0050] According to an embodiment of the present disclosure, balance can characterize the number of labels in each category, and category imbalance may cause the model to perform poorly in some aspects. A larger initial text set can provide more information, but also requires more computing resources.

[0051] According to an embodiment of the present disclosure, timeliness can characterize the efficiency of the initial text set in executing a training task related to a specified scenario for a large model.

[0052] According to an embodiment of the present disclosure, the expected training effect may include the training time, training energy consumption, generalization ability, computational efficiency, accuracy, etc. of performing a specified large model training task using an initial text set.

[0053] According to an embodiment of the present disclosure, a preset tag system can be constructed based on the tag attributes of each of a plurality of preset tags, and the attribute relationship between a plurality of preset tags. Tag attributes can be used to enrich the semantic content of preset tags to improve the comprehensiveness and accuracy of preset tags. Tag attributes may include, for example: tag granularity, tag hierarchy, tag category, tag weight, etc. Those skilled in the art may design the structure or construction method of the preset tag system according to actual needs and application scenarios, etc., and are not limited here. For example, the structure of the preset tag system and the organization method of multiple preset tags can be determined based on training requirements and prior knowledge to construct a preset tag system. For example, prompt words can be used to guide a large model to mine and organize multiple preset tags to construct a preset tag system.

[0054] According to an embodiment of the present disclosure, the tag granularity can characterize the roughness or detail of the preset tag. The tag granularity can include, for example, coarse granularity and fine granularity. The fine-grained preset tag can provide more specific information, and the coarse-grained preset tag can provide a wider range of information. Exemplarily, football skills (coarse granularity), shooting (fine granularity). The tag hierarchy can characterize the hierarchical relationship between the preset tags, so as to facilitate the construction of the preset tag system. The tag hierarchy can include, for example, the top layer, the middle layer and the bottom layer. Exemplarily, sports (top layer)> ball sports (middle layer)> football sports (bottom layer). The tag category can characterize the field, theme or scene involved in the tag, and the tag category can include, for example, the field of mathematics, the field of finance, the question and answer scene, the commodity recommendation scene, the theme of animation works, etc. The same tag category can include multiple preset tags of different granularity or multiple different levels, and a preset tag can also involve multiple different tag categories. Exemplarily, the sports field can include physical education, track and field sports, basketball, dunks, etc., and basketball can involve sports classification, commodity recommendation classification, etc. The tag weight may represent the relative importance of a preset tag, and the tag weight may include, for example, a weight coefficient.

[0055] According to an embodiment of the present disclosure, preset tags with the same or related tag attributes can be clustered according to their respective tag attributes to construct a preset tag system, and the clustering information can characterize the attribute relationship between the multiple preset tags.

[0056] According to an embodiment of the present disclosure, a preset tag system can be constructed based on the tag attributes of each of the multiple preset tags and the attribute relationship between the multiple preset tags. Exemplarily, the attribute relationship may include a hierarchical relationship, an association relationship, a dependency relationship (e.g., the attribute information of one preset tag may depend on the attribute information of another preset tag), etc. between the multiple preset tags.

[0057] Exemplarily, the hierarchical relationship may be used to define the belonging relationship or inclusion relationship between preset tags, etc. The hierarchical relationship may be determined, for example, according to the tag hierarchies of the respective preset tags.

[0058] Exemplarily, the association relationship can be used to define the logic, semantics or connection between preset tags. The association relationship can include, for example, causal relationship, opposition relationship, similarity or correlation relationship, parallel relationship, sequence relationship, combination relationship, etc.

[0059] According to an embodiment of the present disclosure, the preset tag system can be constructed based on a tree structure, a mesh structure, etc., which is not specifically limited here. As an example, the tree structure can be used to construct a preset tag system based on the hierarchical relationship between multiple preset tags. As another example, the mesh structure can be used to construct a preset tag system based on the association relationship between multiple preset tags.

[0060] According to an embodiment of the present disclosure, a preset label system can be determined based on a large model processing a historical text set, and the historical text set can include multiple historical texts. The large model processing of the historical text set can be controlled by using prompt words, and the prompt words can indicate the content of the preset label, the attribute information of the preset label, and the construction method of the preset label system, so that the automatic mining of the preset label and the automatic construction of the preset label system can be realized based on the large model, which is conducive to improving the accuracy and comprehensiveness of the preset labels and the construction efficiency of the preset label system.

[0061] According to an embodiment of the present disclosure, multi-dimensional and multi-granularity labels related to the historical text set can be generated based on the large model and instructions, and the preset label system can be verified and completed by expert combing, etc., aiming to cover preset labels of different granularities, so as to form a good mapping relationship with the historical text set. In order to ensure the consistency and standardization of the preset label system and avoid the appearance of redundant similar labels, the label embedding model can also be trained, and clustering algorithms such as Kmeans can be used to achieve a unified expression (normalization) of synonymous labels, so as to realize the automatic construction of the preset label model and efficiently mine widely covered, high-quality, and normalized preset labels from the historical text set.

[0062] According to the embodiments of the present disclosure, by processing the historical text set through a large model to obtain a preset label system, it is possible to use the large model to fully mine the preset label content and preset label attribute information of the historical text set, and build a more accurate, comprehensive and multi-dimensional preset label system, which provides a basis for the subsequent determination of labels related to the initial text set based on the large model and attribute detection of the initial text set. By using the large model to process the initial text set and the prompt words determined based on the preset label system, the large model can be guided to output labels related to the initial text set, and the use of prompt words to control the large model can be based on the understanding of the initial text set. Further mine the semantic attributes of the initial text set, more accurately and comprehensively obtain the labels related to the initial text set, thereby improving the annotation precision and accuracy of the initial text set. By detecting the attribute evaluation information of the initial text set based on the label, it is possible to accurately evaluate the training effect of the initial text set to perform the specified training task, and improve the training effect of the subsequent use of the initial text set to perform the specified model training task.

[0063] According to an embodiment of the present disclosure, attribute detection is performed on an initial text set based on a label to obtain attribute evaluation information, including: determining task complexity information of the label, wherein the task complexity information characterizes at least one of the execution accuracy and execution efficiency of the initial text set corresponding to the label in performing a specified large model training task; based on the task complexity information, determining complexity evaluation information of the initial text set, the attribute evaluation information including the complexity evaluation information.

[0064] According to an embodiment of the present disclosure, the task complexity information of a label may characterize at least one of the execution accuracy and execution efficiency of the initial text set corresponding to the label in executing the specified large model training task. As an example, the task complexity information may characterize the execution accuracy of the initial text set in executing the specified large model training task. As another example, the task complexity information may characterize the execution efficiency of the initial text set in executing the specified large model training task. As another example, the task complexity information may characterize the execution accuracy and execution efficiency of the initial text set in executing the specified large model training task.

[0065] According to an embodiment of the present disclosure, the task complexity information of the label can characterize the complexity of the model training task performed based on the initial text set. For example, the higher the task complexity of the initial text set, the lower the execution efficiency of the initial text set in performing the specified large model training task, and vice versa. For example, the higher the task complexity of the initial text set, the lower the execution accuracy of the initial text in performing the specified large model training task, and vice versa.

[0066] It should be noted that the task complexity information of the label not only focuses on the lexical complexity and syntactic complexity of the initial text set, but also focuses on the complexity of the large model training task performed by the initial text set. As an example, a large number of advanced mathematics proof tasks are still simple in expression, such as "proving that any even number greater than 2 can be written as the sum of two prime numbers", but it is the Goldbach conjecture with extremely high task complexity.

[0067] According to an embodiment of the present disclosure, the attribute evaluation information of the initial text set may include complexity evaluation information. The complexity evaluation information of the initial text set may be determined based on the task complexity information of the label corresponding to the initial text set. The complexity evaluation information of the initial text set may characterize the expected training effect of the initial text set in performing the specified large model training task. Based on the expected training effect of the initial text set in performing the specified large model training task, a training strategy related to the initial text set may be determined, such as determining the computing resources required for training. Exemplarily, an initial text set with higher complexity may require more computing resources.

[0068] According to an embodiment of the present disclosure, a large model can be used to perform task complexity detection on an initial text set, obtain task complexity information of a label corresponding to the initial text set, and then obtain complexity evaluation information of the initial text set. Exemplarily, the complexity evaluation information of the initial text set can be determined by artificial definition, model fitting, and the like.

[0069] According to one embodiment of the present disclosure, the task complexity information of the label can be set to 1, and the complexity evaluation information of the initial text set can be determined based on the number of labels. For example, the more the number of labels related to the initial text set, the greater the complexity evaluation information of the initial text set, and the more complex the large model training task performed by the initial text set is.

[0070] According to an embodiment of the present disclosure, a difficulty coefficient weight can be set for a tag, and the complexity evaluation information of the initial text set can be determined based on the difficulty coefficient weight of the tag. The task complexity information of the tag can characterize the difficulty coefficient weight of the tag. Exemplarily, the difficulty coefficient weight can be determined based on prior knowledge, semantic information of the tag, dependency of the tag on other tags, etc.

[0071] As an example, the tag corresponding to the initial text set may include one. For example, the task complexity information of the tag may be determined based on the large model, and the task complexity information of the tag may be determined as the complexity evaluation information of the initial text set.

[0072] As another example, the tags corresponding to the initial text set may include multiple tags. For example, the task complexity information of each tag can be determined based on the large model, and the accumulated value of the task complexity of each tag can be determined as the complexity evaluation information of the initial text set. For example, the overall task complexity information of multiple tags can be determined based on the large model, and the overall task complexity information can be determined as the complexity evaluation information of the initial text set.

[0073] According to an embodiment of the present disclosure, the initial text set is associated with a plurality of tags. For example, the tags associated with the initial text set may include N, where N>1 and is a positive integer.

[0074] According to an embodiment of the present disclosure, determining the task complexity information of a label includes: using an initial text set and labels as test data, testing a pre-trained deep learning model to obtain a test indicator; and determining the task complexity information of the label based on the difference between the test indicator and a preset test indicator.

[0075] According to the embodiments of the present disclosure, the test indicators may include accuracy, precision, recall, mean square error (MSE), mean absolute error (MAE), training duration, training energy consumption and other indicators. Those skilled in the art may make reasonable selections based on the training tasks, and no specific limitation is made here.

[0076] According to an embodiment of the present disclosure, the preset test indicator may include a preset threshold value of the test indicator, and the task complexity information of the tag may be determined based on the difference between the test indicator and the preset threshold value of the test indicator.

[0077] For example, the more the accuracy obtained by testing the pre-trained deep learning model is lower than the preset accuracy threshold, the higher the task complexity of the label; the more the accuracy obtained by testing the pre-trained deep learning model is higher than the preset accuracy threshold, the lower the task complexity of the label.

[0078] For example, the more the training time for testing the pre-trained deep learning model exceeds the preset threshold of the training time, the higher the task complexity of the labeling; the more the training time for testing the pre-trained deep learning model is lower than the preset threshold of the training time, the lower the task complexity of the labeling.

[0079] According to an embodiment of the present disclosure, determining task complexity information of a tag includes: determining an associated tag associated with the tag based on an attribute relationship; and performing task complexity detection based on the associated tag to obtain task complexity information.

[0080] According to one embodiment of the present disclosure, the preset label system may adopt a tree structure, and the preset label system may include multiple parent node labels and multiple child node labels, and the multiple parent node labels and the multiple child node labels are connected according to the attribute relationship. The associated label associated with the label may include, for example, the parent node label and / or the child node label of the label in the preset label system, and the task complexity detection may be performed based on the associated label to obtain the task complexity information of the associated label. Among them, the method for determining the task complexity information of the associated label can refer to the method for determining the task complexity information of the label described above, which will not be repeated here.

[0081] According to an embodiment of the present disclosure, the complexity evaluation information of the initial text set can be determined based on the task complexity information of the tag and the associated tag. For example, the cumulative sum of the task complexity information of the tag and the task complexity information of the associated tag can be determined as the complexity evaluation information of the initial text set.

[0082] According to the embodiments of the present disclosure, by determining the associated tags associated with the tags based on the attribute relationship, the tags corresponding to the initial text set can be further expanded and mined, and the completeness and comprehensiveness of the tags can be expanded, thereby improving the annotation accuracy of the initial text set and the accuracy of the attribute evaluation information, which can help to accurately evaluate the expected training effect of the initial text set in performing the specified large model training task.

[0083] According to an embodiment of the present disclosure, the initial text set includes multiple initial texts, at least one initial text is associated with M text tags among N tags, N≥M>1, and N and M are integers.

[0084] According to an embodiment of the present disclosure, the initial text set may include multiple initial texts, the tags related to the initial text set may include multiple text tags, and the multiple text tags may be associated with the multiple initial texts.

[0085] Figure 3 An initial text set and tags related to the initial text set according to an embodiment of the present disclosure are exemplarily shown.

[0086] like Figure 3 As shown, the initial text set may include 4 initial texts, and the labels corresponding to the initial text set may include 6. For example, the first text is associated with 2 text labels, the second text is associated with 3 text labels, the third text is associated with 2 text labels, and the fourth text is associated with 1 text label.

[0087] According to an embodiment of the present disclosure, attribute detection is performed on an initial text set based on labels to obtain attribute evaluation information, including: determining a target text from the initial text, wherein the number of text labels associated with the target text is greater than or equal to a preset label quantity threshold; based on the target text, determining diversity information of the initial text set, the attribute evaluation information including diversity information, and the diversity information indicates the adaptability of the initial text set to a variety of task types, where the task type is the type of large model training task.

[0088] According to an embodiment of the present disclosure, the attribute evaluation information of the initial text set may include diversity information, and the diversity information may indicate the adaptability of the initial text set to a variety of task types. Exemplarily, the task types may include types of large model training tasks, such as customer service conversations, content creation, text generation, machine translation, sentiment analysis, medical field models, financial field models, etc. Based on the adaptability of the initial text set to a variety of task types, a training strategy related to the initial text set may be determined, such as determining the type of training tasks to which the initial text set is applicable. Exemplarily, an initial text set with higher diversity may be more suitable for generalization scenarios, or pre-training of models; an initial text set with lower diversity may be more suitable for single / specific scenarios, or fine-tuning training of models.

[0089] According to an embodiment of the present disclosure, the diversity information of the initial text set can be determined based on the number of tag combinations determined by tags related to the initial text set. For example, the greater the number of tag combinations, the higher the diversity of the initial text set.

[0090] According to another embodiment of the present disclosure, the diversity information of the initial text set can be determined based on the number of tag categories covered by tags related to the initial text set. For example, the more the number of tag categories, the higher the diversity of the initial text set.

[0091] According to an embodiment of the present disclosure, the initial text set may include multiple initial texts, and the initial texts may be associated with one or more tags. Optionally, considering the representativeness of the initial texts, the target texts whose text tags are greater than or equal to a preset tag number threshold may be specifically counted.

[0092] According to the embodiments of the present disclosure, the preset tag quantity threshold can be reasonably set according to actual needs or application scenarios, etc., and is not specifically limited here. For example, the preset tag quantity threshold can be set to 3, and the initial text with the number of associated text tags greater than or equal to 3 can be determined as the target text.

[0093] According to an embodiment of the present disclosure, diversity information of an initial text set can be determined based on a target text. As an example, the diversity of an initial text set can be determined based on the number of target texts, for example, the more target texts there are, the higher the diversity of the initial text set. As another example, the diversity information of an initial text set can be determined based on the number of text tag combinations determined by text tags related to the target text, for example, the more text tag combinations there are, the higher the diversity of the initial text set. As another example, the diversity information of an initial text set can be determined based on the number of text tag categories covered by text tags related to the target text, for example, the more text tag categories there are, the higher the diversity of the initial text set.

[0094] According to an embodiment of the present disclosure, determining the diversity information of the initial text set based on the target text includes: determining the diversity information of the initial text set based on the text distribution of the target text in the initial text set, wherein a plurality of initial texts in the initial text set have a context position relationship, and the text distribution represents the context position of the target text in the initial text set.

[0095] According to an embodiment of the present disclosure, the context position of the target text in the initial text set may include the position, order, paragraph structure, etc. of the target text. Exemplarily, the context position may include the starting part, the middle part, the ending part, the XXth paragraph, the XXth line, the XXth word in the XXth sentence, etc.

[0096] According to an embodiment of the present disclosure, the diversity information of the initial text set can be determined according to the distribution of the target text determined based on the context position. For example, the more comprehensive the target text distribution is, the higher the diversity of the initial text set is. For example, the more evenly the target text is distributed, the higher the diversity of the initial text set is.

[0097] According to an embodiment of the present disclosure, the diversity information of the initial text set can be determined according to the number of target texts distributed in a specific context position. For example, the more target texts distributed in the beginning part, the higher the diversity of the initial text set. For example, the more target texts distributed in the ending part, the higher the diversity of the initial text set.

[0098] According to an embodiment of the present disclosure, the initial text set may include one or more initial texts, and each initial text may correspond to one or more text tags. The attribute evaluation information of the initial text may be determined based on the text tag corresponding to the initial text. The attribute evaluation information of the initial text may, for example, include complexity evaluation information and diversity information of the initial text. Exemplarily, the attribute evaluation information of the initial text set may be determined based on the attribute evaluation information of the initial text, for example, the cumulative sum of the attribute evaluation information of each of the multiple initial texts may be determined as the attribute evaluation information of the initial text set.

[0099] According to an embodiment of the present disclosure, the text processing method based on a large model also includes: in response to a data acquisition request, determining a request tag related to the data acquisition request; based on the matching result between the request tag and the tag, determining a training text matching the request tag from at least one initial text set; and based on the attribute evaluation information of the training text, determining a target training strategy representing the training of the large model to be trained.

[0100] According to an embodiment of the present disclosure, a data acquisition request may characterize the training intent of a large model training task. Exemplarily, the training intent may include the type of a specified large model training task and the expected training effect of the specified model training task. The specified large model training task may, for example, include a specified training task for a large model to be trained.

[0101] According to an embodiment of the present disclosure, a request tag associated with the model training intent can be determined based on the model training intent represented by the data acquisition request. For example, the request tag may include the application field of the model, the application scenario of the model, the function of the model, the training effect of the model, etc. Exemplarily, the request tag may include one or more. The request tag may correspond to a preset tag in a preset tag system, or may correspond to a tag related to the initial text set.

[0102] As an example, a large model and instructions may be used to determine request tags associated with a data acquisition request. The instructions may indicate the model training intent and tags associated with an initial text set.

[0103] According to an embodiment of the present disclosure, a request tag and a tag related to an initial text set may be matched to obtain a matching result. A training text matching the request tag may be determined from at least one initial text set based on the matching result between the request tag and the tag. The training text may be applicable to a model training task that meets the model training intention. Exemplarily, the training text may include one or more.

[0104] According to an embodiment of the present disclosure, the training text may correspond to the initial text in the initial text set. The attribute evaluation information of the training text may be determined based on the attribute evaluation information of the initial text. Exemplarily, the attribute evaluation information of the training text may include complexity evaluation information and diversity information of the training text.

[0105] According to the embodiments of the present disclosure, a target training strategy for training a large model to be trained can be determined based on the attribute evaluation information of the training text. Based on the target training strategy, the training text can be more effectively used to train the large model to be trained, so as to better meet the specific model training intention, which is conducive to improving the training efficiency and performance of the model.

[0106] According to one embodiment of the present disclosure, the target training strategy may include the selection of training texts, the order in which training texts are used, the computing resources required for training, the applicable scenarios of training texts, etc. For example, for fine-tuning training tasks, training texts with high complexity can be selected to improve the performance of the model on specific tasks. For example, training texts with low complexity can be used for pre-training first, and then training texts with high complexity can be used to iterate and optimize the model. For example, the number and performance of GPUs required for training can be determined based on the complexity evaluation information of the training text. For example, training texts with high diversity can be used for generalization training tasks, and training texts with low diversity can be used for single-scenario training tasks.

[0107] Figure 4 A schematic diagram of determining a target training strategy according to an embodiment of the present disclosure is exemplified.

[0108] According to an embodiment of the present disclosure, the training text may include multiple training texts, and the training texts may have complexity evaluation information. A complexity preset threshold may be set, and the training order of each training text may be determined based on the complexity preset threshold and the complexity evaluation information of each training text.

[0109] like Figure 4 As shown, the first training text 401 whose complexity evaluation information is lower than the preset complexity threshold can be used in the pre-training stage of the model, and the second training text 402 whose complexity evaluation information is higher than the preset complexity threshold can be used in the iterative training stage of the model.

[0110] According to an embodiment of the present disclosure, a large model-based data processing system is also provided.

[0111] Figure 5 A large model-based data processing system according to an embodiment of the present disclosure is schematically shown.

[0112] like Figure 5 As shown, the big model-based data processing system 500 may include a preset label system automatic construction subsystem 501, a data marking subsystem 502, a data attribute detection subsystem 503 and a data training subsystem 504.

[0113] According to an embodiment of the present disclosure, the preset tag system automatic construction subsystem 501 can be used to process the historical text set, fully mine and organize multiple preset tags, and construct a preset tag system based on multiple preset tags and attribute relationships between multiple preset tags.

[0114] According to an embodiment of the present disclosure, the data labeling subsystem 502 may be used to process the initial text set and the prompt words determined based on a preset labeling system, label the initial text set, and obtain labels related to the initial text set.

[0115] According to an embodiment of the present disclosure, the data attribute detection subsystem 503 can be used to perform attribute detection on the initial text set to obtain attribute evaluation information, which can characterize the expected training effect of performing a specified large model training task based on the initial text set.

[0116] According to an embodiment of the present disclosure, the data training subsystem 504 may be used to determine training texts that match the requested tags, so that a target training strategy that represents training of the large model to be trained may be determined based on the attribute evaluation information of the training texts.

[0117] The data processing system based on the big model provided by the embodiment of the present disclosure can automatically build a relatively complete preset label system, and can efficiently and directionally expand the preset label system by enriching and optimizing the preset labels, so as to support label mining and automatic annotation of the initial text set based on the preset label system, thereby driving the data production and iteration of downstream tasks. In addition, the data processing system based on the big model can perform attribute detection on the data from multiple dimensions, so as to determine the target training strategy based on the training intention and attribute evaluation information, thereby helping to improve the training efficiency and performance of the model.

[0118] It should be noted that the above embodiments of the present disclosure are only exemplary embodiments, but are not limited thereto, and may also include other large model-based text processing methods known in the art, as long as they can improve the training effect of the specified large model training task.

[0119] According to an embodiment of the present disclosure, a text processing device based on a large model is also provided.

[0120] Figure 6 A block diagram of a text processing apparatus based on a large model according to an embodiment of the present disclosure is schematically shown.

[0121] like Figure 6 As shown, the large model-based text processing apparatus 600 may include a determination module 610 and a detection module 620 .

[0122] A determination module 610 is used to respond to the data evaluation request, use the big model to process the initial text set and the prompt words determined based on the preset label system, and obtain labels related to the initial text set, wherein the preset label system includes multiple preset labels and attribute relationships between the multiple preset labels, and the preset label system is determined based on the big model processing the historical text set;

[0123] The detection module 620 is used to perform attribute detection on the initial text set based on the label to obtain attribute evaluation information, where the attribute evaluation information represents the expected training effect of executing the specified large model training task based on the initial text set.

[0124] According to an embodiment of the present disclosure, the detection module 620 may include a first determination submodule and a second determination submodule.

[0125] The first determination submodule is used to determine task complexity information of the label, wherein the task complexity information represents at least one of the execution accuracy and execution efficiency of the initial text set corresponding to the label to perform the specified large model training task.

[0126] The second determination submodule is used to determine complexity evaluation information of the initial text set based on the task complexity information, and the attribute evaluation information includes complexity evaluation information.

[0127] According to an embodiment of the present disclosure, the initial text set is associated with a plurality of tags; the first determination submodule may include a testing unit and a first determination unit.

[0128] The testing unit is used to test the pre-trained deep learning model using the initial text set and labels as test data to obtain test indicators.

[0129] The first determining unit is used to determine the task complexity information of the label based on the difference between the test indicator and the preset test indicator.

[0130] According to an embodiment of the present disclosure, the first determining submodule may include an associated tag determining unit and a second determining unit.

[0131] The associated tag determining unit is used to determine an associated tag associated with the tag based on the attribute relationship.

[0132] The second determining unit is used to perform task complexity detection based on the associated tags to obtain task complexity information.

[0133] According to an embodiment of the present disclosure, the initial text set includes multiple initial texts, at least one initial text is associated with M text tags among N tags, N≥M>1, and N and M are integers; wherein the detection module may include a third determination submodule and a fourth determination submodule.

[0134] The third determination submodule is used to determine the target text from the initial text, and the number of text tags associated with the target text is greater than or equal to a preset tag number threshold.

[0135] The fourth determination submodule is used to determine the diversity information of the initial text set based on the target text. The attribute evaluation information includes diversity information. The diversity information indicates the adaptability of the initial text set to multiple task types. The task type is the type of large model training task.

[0136] According to an embodiment of the present disclosure, the fourth determining submodule may include a third determining unit.

[0137] The third determination unit is used to determine the diversity information of the initial text set based on the text distribution of the target text in the initial text set, wherein a plurality of initial texts in the initial text set have a context position relationship, and the text distribution represents the context position of the target text in the initial text set.

[0138] According to an embodiment of the present disclosure, the large model-based text processing apparatus may further include an acquisition module, a matching module, and a training strategy module.

[0139] The acquisition module is used to determine a request tag related to the data acquisition request in response to the data acquisition request.

[0140] The matching module is used to determine training text matching the request tag from at least one initial text set based on the matching result between the request tag and the tag.

[0141] The training strategy module is used to determine a target training strategy for training a large model to be trained based on attribute evaluation information of the training text.

[0142] It should be noted that the text processing device part based on the big model in the embodiment of the present disclosure corresponds to the text processing method part based on the big model in the embodiment of the present disclosure. The description of the text processing device part based on the big model specifically refers to the text processing method part based on the big model, which will not be repeated here.

[0143] An embodiment of the present disclosure also provides an AI (Artificial Intelligence) agent configured to execute the large model-based text processing method as described above.

[0144] Figure 7 The structural block diagram of an artificial intelligence agent according to an embodiment of the present disclosure is schematically shown.

[0145] In the embodiments of the present disclosure, inspired by the von Neumann structure in modern computer theory, such as Figure 7 As shown, the AI ​​agent 700 may include five core modules: an input module 710 , a control module 720 , a storage module 730 , a calculation module 740 and an output module 750 .

[0146] The input module 710 is responsible for receiving or sensing information such as queries, requests, instructions, signals or data from the outside world (such as users or external environment), and converting them into a format that can be understood and processed by the AI ​​agent 700. The input module 710 is the primary link for the AI ​​agent 700 to interact with the outside world. It enables the AI ​​agent 700 to efficiently and accurately obtain the necessary "sensory" information from the outside world and respond to this information.

[0147] In an example, the input module 710 may input the data evaluation request described above and receive the data acquisition request.

[0148] In the example, the control module 720 is the core support for the AI ​​agent 700 to process complex tasks. The control module 720 can execute the large model-based text processing method described above.

[0149] In the example, the control module 720 will continuously interact with the storage module 730, the computing module 740 and / or the output module 750 during operation. However, it should be noted that in the embodiment of the present disclosure, the control module 720 acts as a single initiator to initiate communication with the storage module 730, the computing module 740 and / or the output module 750, and there is no communication coupling between the storage module 730, the computing module 740 and the output module 750.

[0150] In the example, the performance of the control module 720 may be closely related to the large model on which the AI ​​agent 700 is based. In order to give full play to the capabilities of the large language model, the internal structure of the control module 720 may be designed to be highly configurable and extensible to cope with various types of tasks and requirements in real scenarios.

[0151] The storage module 730 may be responsible for memorizing information such as historical conversations, event flows, etc. The preset tag system and attribute evaluation information as described above may be included in the storage module 730 .

[0152] In the example, after the AI ​​agent 700 obtains the data evaluation request, the AI ​​agent 700 can process the initial text set and the prompt words determined based on the preset label system. The preset label system can be stored in the storage module 730. The AI ​​agent 700 can retrieve the labels related to the initial text set from the storage module 730 and feed them back to the control module 720. Then, the control module 720 can use the fed-back labels to perform attribute detection on the initial text set, obtain attribute evaluation information, and pass the attribute evaluation information to the output module 750.

[0153] The operation module 740 may be viewed as a predefined tool library. An operator for determining the task complexity and complexity evaluation information as described above may be included in the operation module 740 .

[0154] In the example, when the AI ​​agent 700 needs to determine the complexity evaluation information of the initial text set, the relevant operator can be called from the operation module 740 and fed back to the control module 720. Then, the control module 720 can use the operator fed back to obtain the complexity evaluation information of the initial text set, and pass the complexity evaluation information to the output module 750. It can be understood that although the large language model has excellent language understanding and generation capabilities, it is the same as a human being. Without the help of any tools, the tasks that can be solved are very limited. When the AI ​​agent 700 is given the ability to call tools, it can achieve tasks such as completing mathematical operations with the help of a calculator, completing data analysis with the help of python, and completing weather forecasts with the help of search engines.

[0155] In an example, the output module 750 may output the property evaluation information and the target training strategy described above.

[0156] The AI ​​agent 700 according to the embodiment of the present disclosure can simply and effectively improve the level of intelligence, and improve flexibility and versatility.

[0157] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0158] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.

[0159] According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the method as described above.

[0160] According to an embodiment of the present disclosure, a computer program product includes a computer program, and when the computer program is executed by a processor, the computer program implements the method as described above.

[0161] Figure 8 A block diagram of an electronic device suitable for implementing a large model-based text 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.

[0162] like Figure 8 As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0163] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0164] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 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 801 performs the various methods and processes described above, such as a text processing method based on a large model. For example, in some embodiments, the text processing method based on a large model may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the text processing method based on the large model described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the large model-based text processing method in any other appropriate manner (eg, by means of firmware).

[0165] 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), load 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.

[0166] 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 text processing device, so that the program code, when executed by the processor or controller, causes 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.

[0167] 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.

[0168] 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).

[0169] 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.

[0170] 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.

[0171] 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.

[0172] 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 text processing method based on a large model, comprising: In response to the data evaluation request, the initial text set and the prompt words determined based on the preset label system are processed by the large model to obtain labels related to the initial text set, wherein the preset label system includes a plurality of preset labels and attribute relationships between the plurality of preset labels, and the preset label system is determined based on the large model processing the historical text set; Attribute detection is performed on the initial text set based on the label to obtain attribute evaluation information, where the attribute evaluation information represents an expected training effect of performing a specified large model training task based on the initial text set.

2. The method according to claim 1, wherein: The performing attribute detection on the initial text set based on the label to obtain attribute evaluation information includes: Determine task complexity information of the label, wherein the task complexity information represents at least one of an execution accuracy and an execution efficiency of the initial text set corresponding to the label in executing the specified large model training task; Based on the task complexity information, complexity evaluation information of the initial text set is determined, and the attribute evaluation information includes the complexity evaluation information.

3. The method according to claim 2, wherein: The initial text set is associated with a plurality of the tags; The step of determining the task complexity information of the tag includes: Using the initial text set and the labels as test data, testing the pre-trained deep learning model to obtain a test index; and Based on the difference between the test indicator and a preset test indicator, the task complexity information of the label is determined.

4. The method according to claim 2, wherein: The task complexity information of determining the label includes: Determining an associated tag associated with the tag based on the attribute relationship; and The task complexity is detected based on the associated tags to obtain the task complexity information.

5. The method according to claim 1, wherein the initial text set comprises a plurality of initial texts, at least one of the initial texts is associated with M text tags among the N tags, N≥M>1, and N and M are integers; in, The performing attribute detection on the initial text set based on the label to obtain attribute evaluation information includes: Determining a target text from the initial text, wherein the number of text tags associated with the target text is greater than or equal to a preset tag number threshold; Based on the target text, diversity information of the initial text set is determined, the attribute evaluation information includes the diversity information, and the diversity information indicates the adaptability of the initial text set to multiple task types, and the task type is the type of large model training task.

6. The method according to claim 5, wherein: Determining the diversity information of the initial text set based on the target text includes: Based on the text distribution of the target text in the initial text set, diversity information of the initial text set is determined, wherein a plurality of initial texts in the initial text set have context position relationships, and the text distribution represents the context position of the target text in the initial text set.

7. The method according to claim 1, further comprising: In response to a data acquisition request, determining a request tag associated with the data acquisition request; Based on the matching result between the request tag and the tag, determining a training text matching the request tag from at least one of the initial text sets; as well as Based on the attribute evaluation information of the training text, a target training strategy representing training of the large model to be trained is determined.

8. An artificial intelligence agent configured to execute the method according to any one of claims 1-7.

9. A text processing device based on a large model, comprising: A determination module, configured to, in response to a data evaluation request, use a large model to process an initial text set and prompt words determined based on a preset label system to obtain labels related to the initial text set, wherein the preset label system includes a plurality of preset labels and attribute relationships between the plurality of preset labels, and the preset label system is determined based on the large model processing the historical text set; A detection module is used to perform attribute detection on the initial text set based on the label to obtain attribute evaluation information, wherein the attribute evaluation information represents the expected training effect of performing a specified large model training task based on the initial text set.

10. The device according to claim 9, wherein: The detection module comprises: A first determination submodule is used to determine task complexity information of the label, wherein the task complexity information represents at least one of an execution accuracy and an execution efficiency of the initial text set corresponding to the label to perform the specified large model training task; The second determination submodule is used to determine complexity evaluation information of the initial text set based on the task complexity information, and the attribute evaluation information includes the complexity evaluation information.

11. The device according to claim 10, wherein: The initial text set is associated with a plurality of the tags; Wherein, the first determining submodule includes: A testing unit, configured to test the pre-trained deep learning model using the initial text set and the label as test data to obtain a test index; and The first determining unit is used to determine the task complexity information of the label based on the difference between the test indicator and the preset test indicator.

12. The device according to claim 10, wherein: The first determining submodule includes: an associated tag determining unit, configured to determine an associated tag associated with the tag based on the attribute relationship; and The second determining unit is used to perform task complexity detection based on the associated tag to obtain the task complexity information.

13. The apparatus according to claim 9, wherein the initial text set comprises a plurality of initial texts, at least one of the initial texts is associated with M text tags among the N tags, N≥M>1, and N and M are integers; in, The detection module comprises: A third determination submodule is used to determine a target text from the initial text, wherein the number of text tags associated with the target text is greater than or equal to a preset tag number threshold; The fourth determination submodule is used to determine the diversity information of the initial text set based on the target text, the attribute evaluation information includes the diversity information, and the diversity information indicates the adaptability of the initial text set to multiple task types, and the task type is the type of large model training task.

14. The device according to claim 13, wherein: The fourth determining submodule includes: The third determination unit is used to determine the diversity information of the initial text set based on the text distribution of the target text in the initial text set, wherein a plurality of initial texts in the initial text set have a context position relationship, and the text distribution represents the context position of the target text in the initial text set.

15. The apparatus according to claim 9, further comprising: an acquisition module, configured to determine, in response to a data acquisition request, a request tag associated with the data acquisition request; A matching module, configured to determine, from at least one of the initial text sets, a training text matching the request tag based on a matching result between the request tag and the tag; as well as The training strategy module is used to determine a target training strategy for training the large model to be trained based on the attribute evaluation information of the training text.

16. An electronic device comprising: 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 enable the at least one processor to perform the method according to any one of claims 1 to 7.

17. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

18. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images