Plug-in calling method based on large model, electronic equipment and storage medium

Abstract

The invention relates to the technical field of large language models, in particular to a plug-in calling method based on a large model, electronic equipment and a storage medium, the method comprises the following steps: matching a preset plug-in keyword in a preset plug-in keyword library with any request keyword in a target request text input by a user, and directly determining a target plug-in or determining a plurality of alternative plug-ins according to the matched preset plug-ins, and when the request keyword is not matched with the preset plug-ins, taking all the preset plug-ins as the alternative plug-ins, and screening out the target plug-ins from the plurality of alternative plug-ins through a given large language model. Extracting target parameter information from the target request text, sending the target parameter information to a target plug-in to obtain a plug-in feedback result, and taking the plug-in feedback result as a request feedback text when a judgment result of the plug-in feedback result is that a preset requirement is met; according to the method and the device, the accuracy of selecting the target plug-in in the dialogue system can be improved, and the selected target plug-in better meets the requirements of a user.

Description

Technical Field

[0001] The present invention relates to the technical field of large language models, and in particular to a plug-in calling method, electronic device and storage medium based on a large model. Background Art

[0002] The large language model is a language model based on deep learning. It is a model that can automatically learn language rules and generate text through training on a large-scale corpus. After a complete process of pre-training, fine-tuning and reinforcement learning, it has strong capabilities and has been used in human-computer dialogue systems. However, practice has found that it still has some shortcomings. Since the request questions entered by users involve content in multiple fields and the demand scenarios are varied, for example, they want to obtain real-time or future weather information for a city, perform complex mathematical operations, generate formulas, draw charts for display, and screen the required network sources. When encountering these demand scenarios, the response accuracy of the dialogue system is not high, which affects the user experience. Summary of the invention

[0003] In response to the above technical problems, the present invention provides a plug-in calling method, electronic device and storage medium based on a large model, which can improve the accuracy of target plug-in selection in a dialogue system and make the selected target plug-in better meet user needs.

[0004] According to a first aspect of the present invention, there is provided a plug-in calling method based on a large model, comprising the following steps:

[0005] A target request text input by a user to a given large language model is received, and a plurality of request keywords are extracted from the target request text.

[0006] Match each request keyword with the preset plug-in keyword in the preset plug-in keyword library. Based on the keyword matching results, when the number of matched preset plug-ins is 1, the matched preset plug-in is used as the target plug-in; when the number of matched preset plug-ins is greater than 1, determine the alternative plug-in corresponding to each matched request keyword; when no preset plug-in is matched, all preset plug-ins are used as alternative plug-ins.

[0007] A semantic analysis is performed on the target request text and the plug-in description text and the input parameter description text corresponding to each candidate plug-in by using a given large language model. According to the semantic analysis result, a target plug-in is screened out from a number of candidate plug-ins to call the screened target plug-in.

[0008] According to the input parameter description text corresponding to each target plug-in called, the target parameter information is extracted from the target request text through a given large language model, and the target parameter information is converted according to the preset parameter format of the corresponding target plug-in and input into the corresponding target plug-in to obtain the plug-in feedback result.

[0009] The plug-in feedback result is input into a given large language model for judgment. When the judgment result of the plug-in feedback result is that it meets the preset requirements, the plug-in feedback result is output as a request feedback text.

[0010] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which at least one instruction or at least one program is stored, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the above-mentioned large model-based plug-in calling method.

[0011] According to a third aspect of the present invention, there is provided an electronic device comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0012] The present invention has at least the following beneficial effects:

[0013] The present invention provides a plug-in calling method based on a large model. First, preset plug-in keywords in a preset plug-in keyword library are matched with any request keyword in a target request text input by a user. When there is no matching request keyword, all preset plug-ins are used as candidate plug-ins. When there is only one matching preset plug-in, the preset plug-in is used as a target plug-in. When there are multiple matching preset plug-ins, all the matching preset plug-ins are used as candidate plug-ins. Then, for several candidate plug-ins, a large language model performs semantic analysis on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in. The target plug-in is screened out according to the semantic analysis result to call the screened target plug-in. By further screening several target plug-ins according to the content of the target request text and the input parameter description of the target plug-in, the accuracy of target plug-in selection in the dialogue system can be improved, so that the selected target plug-in can better meet the needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 A flowchart of a plug-in calling method based on a large model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0017] The embodiment of the present invention provides a plug-in calling method based on a large model, such as Figure 1 As shown, the method comprises the following steps:

[0018] S100, receiving a target request text input by a user into a given large language model, and extracting a number of request keywords from the target request text. For example, a keyword extraction tool is used to extract request keywords from the target request text.

[0019] Specifically, the target request text includes several rounds of dialogue information texts input by the user in the dialogue system; it can be understood that the target request text includes but is not limited to the current round of information text input by the user in the dialogue system. In the specific implementation, it is also necessary to judge the sensitivity of the request text input by the user and whether the sentence meaning is clear. When sensitive words are involved in the request text, the model directly replies tactfully; when the sentence meaning of the request text is not clear, for example, the request text is "What information is contained in it", then for multi-round dialogues, it is necessary to obtain the text of the context and rewrite the complete information through a given large language model. For single-round dialogues or multi-round dialogues where the complete information cannot be rewritten through the context, a given large language model outputs a preset reply text, such as "Please provide further information" to obtain a complete question with clear meaning. When judging whether the meaning of the request text is clearly expressed, one implementation method is to detect whether the request text contains specified words, such as: it, he, this, that, etc. If so, it is judged that the meaning of the request text is not clearly expressed. Another implementation method is to combine the problems in the actual application scenario to mark whether the meaning is complete and then train a binary classification model to obtain a judgment result with high accuracy and better generalization.

[0020] In one embodiment, the following steps are also included before step S100:

[0021] S101, receiving an initial request text input by a user into a given large language model, and extracting a number of initial keywords from the initial request text.

[0022] S102, traverse several initial keywords, and when there is an initial keyword corresponding to any preset first type keyword, intercept the initial request text; it can be understood that the first type keyword refers to a preset keyword that must be intercepted, such as some highly sensitive words.

[0023] Specifically, the risk levels of the first type of keywords, the second type of keywords, and the third type of keywords decrease in sequence.

[0024] S103, when there is an initial keyword corresponding to any preset second type keyword, obtain the detection result of the preset anomaly detection model on the initial request text, when the detection result is abnormal, intercept the initial request text, when the detection result is non-abnormal but the confidence of the detection result is less than the first preset threshold, intercept the initial request text; it can be understood that: the anomaly detection model is a binary classification model trained in advance to evaluate whether the text is an abnormal text. For example, when the initial keyword in the initial request text is detected, when the initial keyword risk exceeds the preset degree, it is considered to be an abnormal text, otherwise it is a non-abnormal text.

[0025] Furthermore, those skilled in the art may set the first preset threshold value, such as 80%, according to actual needs.

[0026] S104, when there is an initial keyword corresponding to any preset third type keyword, obtain the detection result of the anomaly detection model on the initial request text, and when the detection result is abnormal and the confidence of the detection result is greater than the second preset threshold, intercept the initial request text.

[0027] Furthermore, those skilled in the art may set the second preset threshold value, such as 95%, according to actual needs.

[0028] As mentioned above, when receiving any request text sent by the user, the request text must first be risk assessed and screened. When the risk level is too high, it will be intercepted. When the risk level is not high, the detection results of the anomaly detection model must be comprehensively considered, and the final detection result is determined by voting, which improves the rationality of the judgment on whether the text is abnormal. The text that passes the detection is used as the target request text and subsequent plug-ins are selected and called.

[0029] S200, each request keyword is matched with a preset plug-in keyword in a preset plug-in keyword library. Based on the keyword matching result, when the number of matched preset plug-ins is 1, the matched preset plug-in is used as a target plug-in; when the number of matched preset plug-ins is greater than 1, an alternative plug-in corresponding to each matched request keyword is determined; when no preset plug-in is matched, all preset plug-ins are used as alternative plug-ins; it can be understood that: in the process of determining the alternative plug-in corresponding to each matched request keyword, several target plug-ins corresponding to the matched request keyword are determined from the preset word comparison table.

[0030] Specifically, the preset word comparison table includes several preset plug-ins and several plug-in keywords corresponding to each preset plug-in. Among them, the same plug-in keyword can also correspond to several preset plug-ins. The following examples are given: when the target request text involves keywords corresponding to calculation questions, the preset plug-ins are calculators, python executors, etc.; when the target request text involves keywords corresponding to formula writing, the preset plug-ins are math tools, etc.; when the target request text involves real-time information data, the preset plug-ins are information media plug-ins, etc.; when the target request text involves keywords for drawing statistical pie charts, bar charts, and flow charts, the preset plug-ins are drawing tools, etc.; when the target request text involves weather or earthquake related messages, the preset plug-ins are weather tools. In the specific implementation, one of the tools can be used according to actual needs, or multiple tools can be used in parallel or serially, and customized selection can also be made according to project requirements.

[0031] It should also be noted that for a target request text containing at least one question, the target request text needs to be disassembled and the plug-in needs to be called in steps. For example, if the target request text entered by the user is: Please help me search for the recent stock price of Moutai and draw a line chart. It needs to be divided into two steps: a. Search for the recent stock price of Moutai, b. Draw a line chart, and the plug-ins called respectively are the search tool and the statistical chart drawing tool.

[0032] Furthermore, the method further screens out candidate plug-ins corresponding to each matching request keyword through the following steps:

[0033] S201: For any matched request keyword, when the number of preset plug-ins corresponding to the request keyword is 1, determine the preset plug-in corresponding to the request keyword as a candidate plug-in.

[0034] S202, when the number of preset plug-ins corresponding to the request keyword is greater than 1, obtain the preset interface stability score of each preset plug-in corresponding to the request keyword and the number of request keywords corresponding to each preset plug-in; for example, when the request keyword is calculation, the corresponding preset plug-in has several tools that can realize the calculation function.

[0035] Furthermore, the preset interface stability score of the preset plug-in refers to an artificial score obtained according to the API test result of the preset plug-in. Among them, the API test includes the correctness test and stability test of the API interface.

[0036] S203: Determine the preset plug-in corresponding to the largest number of requested keywords and the preset plug-in corresponding to the second largest number of requested keywords as key plug-ins.

[0037] S204, when D max / D min <η, D max The corresponding preset plug-in is determined as the alternative plug-in, where D max is the value corresponding to the maximum number of requested keywords, D min is the value corresponding to the second most requested keyword quantity, and n is the preset ratio threshold. For example, n can be set to 2 / 3.

[0038] S205, when D max / D min ≥η, the target score corresponding to each preset plug-in is calculated by weighted sum according to the preset score weight, quantity weight, preset interface stability score of each preset plug-in corresponding to the request keyword and the number of request keywords corresponding to each preset plug-in, and the preset plug-in corresponding to the maximum target score is used as an alternative plug-in. In a specific implementation, since the number of keywords corresponding to the preset plug-in is generally in the single digit, the preset interface stability score can be normalized to between 1 and 5.

[0039] In the above, the number of matched keywords is first determined. When no request keyword is matched, all preset plug-ins are used as alternative plug-ins. When a request keyword is matched, based on multiple preset plug-ins corresponding to a request keyword, an alternative plug-in corresponding to a request keyword is screened out by comprehensively considering the interface stability score of each preset plug-in and the corresponding number of request keywords. In this way, the accuracy of the selection of alternative plug-ins is improved, and the selected alternative plug-in is more likely to meet the user's needs.

[0040] S300, semantic analysis is performed on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in by using a given large language model, and based on the semantic analysis result, a target plug-in is screened out from a number of candidate plug-ins to call the screened target plug-in; it can be understood that the plug-in description text is a plug-in introduction text generated based on the functional scope text and the usage method text of the plug-in.

[0041] Specifically, the input parameter description text corresponding to the alternative plug-in refers to a text composed of the input parameter structure description text and the parameter format description text extracted from the preset parameter information required by the alternative plug-in. For example, the input parameter structure description text includes optional parameters and required parameter descriptions; the parameter format description text includes the format requirements of each parameter, for example, the date type parameters need to specify whether their specific format is to month, day, or to hours, minutes, seconds, etc.

[0042] As mentioned above, after obtaining the candidate plug-ins, it is necessary to further screen the several candidate plug-ins obtained above, through the semantic analysis of the target request text content by the large language model, as well as the semantic analysis of the plug-in description text and the input parameter description text, by comparing whether the semantics are consistent, the final required target plug-in is screened out, thereby improving the accuracy and reliability of the plug-in call.

[0043] S400, according to the input parameter description text corresponding to each target plug-in called, extract the target parameter information from the target request text through a given large language model, and convert the target parameter information into a preset parameter format of the corresponding target plug-in and input it into the corresponding target plug-in to obtain the plug-in feedback result. For example, including but not limited to outputting the obtained target parameter information in JSON format.

[0044] In a specific embodiment, step S400 further includes the following steps:

[0045] S401, verifying the parameter format of target parameter information converted according to a preset parameter format according to the parameter format description text corresponding to the target plug-in.

[0046] Furthermore, the target parameter information corresponding to the target plug-in refers to corresponding parameter information extracted from the target request text according to the input parameter description text of the target plug-in.

[0047] S402, when the verification is consistent, the converted parameters are input into the corresponding target plug-in through the corresponding API interface to obtain the output result of the target plug-in.

[0048] S403, when the verification is inconsistent, the obtained parameter format is corrected according to the parameter format description text corresponding to the target plug-in to obtain new parameters, and the new parameters are input into the corresponding target plug-in through the corresponding API interface to obtain the output result of the target plug-in.

[0049] S404, combining and processing the output results of all target plug-ins to obtain plug-in feedback results; this can be understood as: according to the output result of each target plug-in, post-processing all the output results and sending them to the large language processing model.

[0050] As described above, the format of the parameters is checked before the target parameter information is input into the target plug-in to prevent the situation where the target plug-in cannot be successfully input into the target plug-in or the target plug-in cannot output accurate results due to format problems, thereby ensuring the normal use of the target plug-in and accurate output of data.

[0051] S500, input the plug-in feedback result into a given large language model for judgment. When the judgment result of the plug-in feedback result meets the preset requirements, the plug-in feedback result is output as a request feedback text. It can be understood that: the given large language model determines whether the plug-in feedback result can answer the user's question through the reflection function. When it can answer, the large language model sorts and organizes the combined results, that is, the feedback results of multiple plug-ins are written into the prompt together for the given large language model to make the final output, and the feedback result of the current round of questions is obtained and displayed on the interface of the dialogue system. In another embodiment, the target request text and the corresponding plug-in feedback result can also be sent to the base model together, and the base model and the answerable prompt are combined to determine whether the plug-in feedback result can answer the user's question. When the base model cannot make a judgment, it is necessary to construct the instruction data selected by the plug-in as a training set and a test set to fine-tune the base model. Each constructed instruction data is the request prompt and the ideal output result that are actually input into a given large language model during actual use. For example, the output result at least includes the selected plug-in and plug-in parameters. In specific implementation, the plug-in parameters need to be clearly defined in combination with the characteristics of each plug-in. For example, search tools tend to use keywords for searches and expand as many questions as possible.

[0052] Furthermore, the method further comprises the following steps:

[0053] S001, calculating the text similarity between the received target request text and each historical request text in the historical plug-in selection library; this can be understood as: storing the plug-in selection results of the historical request questions in the historical plug-in selection library accordingly.

[0054] S002, when the maximum text similarity exceeds the preset similarity threshold, the historical plug-in selection result corresponding to the maximum text similarity is sent to the given large language model, so that the given large language model determines the plug-in corresponding to the historical plug-in selection result as the target plug-in; it can be understood that: when the maximum text similarity exceeds the preset similarity threshold, it indicates that the current request question has completed the plug-in selection, and its selection result is introduced into the plug-in selection prompt as a few shots to guide the model to give priority to this result.

[0055] Furthermore, the method further comprises the following steps:

[0056] S10, when the judgment result of the plug-in feedback result is that it does not meet the preset requirements, construct a training data set corresponding to each preset plug-in and input it into a given large language model to obtain a model output result; the training data set refers to a data set obtained after expanding the input of the sample and adding prompt words; it can be understood as: adding sample data in actual business scenarios and adding prompt words in different business fields.

[0057] S20, adjusting the given large language model according to the model output results and the actual results in the training data set to obtain a trained target large language model; those skilled in the art are aware of the specific process of training the large language model according to the data set, which will not be repeated here.

[0058] S30, taking the target large language model as a new given large language model, and returning to the execution step of performing semantic analysis on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in through the given large language model, and screening out the target plug-in from a number of candidate plug-ins according to the semantic analysis result, until the plug-in feedback result is judged to meet the preset requirements.

[0059] As mentioned above, when the request feedback result does not meet the requirements, it indicates that the accuracy of the large language model is not enough. Therefore, it is necessary to increase the sample data in the actual scenario, further train the large language model through sample expansion input, complete the update of the large language model, and improve the accuracy of the model output results.

[0060] In summary, the present invention provides a plug-in calling method based on a large model. First, the preset plug-in keywords in the preset plug-in keyword library are matched with any request keyword in the target request text input by the user. When there is no matching request keyword, all preset plug-ins are used as candidate plug-ins. When there is only one matching preset plug-in, the preset plug-in is used as the target plug-in. When the number of matched preset plug-ins is multiple, all the matched preset plug-ins are used as candidate plug-ins. Then, for several candidate plug-ins, the large language model performs semantic analysis on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in. The target plug-in is filtered out according to the semantic analysis result to call the filtered target plug-in. By further filtering several target plug-ins according to the content of the target request text and the input parameter description of the target plug-in, the accuracy of target plug-in selection in the dialogue system can be improved, so that the selected target plug-in can better meet the needs of users.

[0061] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0062] An embodiment of the present invention further provides an electronic device, comprising a processor and the aforementioned non-transitory computer-readable storage medium.

[0063] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will also be appreciated by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A plug-in calling method based on a large model, characterized in that: The method comprises the following steps: receiving a target request text input by a user into a given large language model, and extracting a plurality of request keywords from the target request text; Match each request keyword with the preset plug-in keyword in the preset plug-in keyword library. Based on the keyword matching results, when the number of matched preset plug-ins is 1, the matched preset plug-in is used as the target plug-in; when the number of matched preset plug-ins is greater than 1, determine the candidate plug-in corresponding to each matched request keyword; when no preset plug-in is matched, all preset plug-ins are used as candidate plug-ins; Perform semantic analysis on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in by using a given large language model, and select a target plug-in from a number of candidate plug-ins according to the semantic analysis result, so as to call the selected target plug-in; According to the input parameter description text corresponding to each target plug-in called, the target parameter information is extracted from the target request text through a given large language model, and the target parameter information is converted according to the preset parameter format of the corresponding target plug-in and input into the corresponding target plug-in to obtain the plug-in feedback result; The plug-in feedback result is input into a given large language model for judgment. When the judgment result of the plug-in feedback result is that it meets the preset requirements, the plug-in feedback result is output as a request feedback text.

2. The plug-in calling method based on a large model according to claim 1, characterized in that: Before step S100, the following steps are also included: receiving an initial request text input by a user into a given large language model, and extracting a number of initial keywords from the initial request text; Traversing a number of initial keywords, when there is an initial keyword corresponding to any preset first type keyword, intercepting the initial request text; When there is an initial keyword corresponding to any preset second type keyword, obtaining a detection result of the preset anomaly detection model on the initial request text, and when the detection result is abnormal, intercepting the initial request text; when the detection result is non-abnormal but the confidence of the detection result is less than a first preset threshold, intercepting the initial request text; When there is an initial keyword corresponding to any preset third type keyword, the detection result of the anomaly detection model on the initial request text is obtained, and when the detection result is abnormal and the confidence of the detection result is greater than the second preset threshold, the initial request text is intercepted.

3. The plug-in calling method based on a large model according to claim 1, characterized in that: The method further determines the candidate plug-in corresponding to each matching request keyword through the following steps: For any matching request keyword, when the number of preset plug-ins corresponding to the request keyword is 1, the preset plug-in corresponding to the request keyword is determined as a candidate plug-in; When the number of preset plug-ins corresponding to the request keyword is greater than 1, the preset interface stability score of each preset plug-in corresponding to the request keyword and the number of request keywords corresponding to each preset plug-in are obtained; Determine the preset plug-in corresponding to the largest number of requested keywords and the preset plug-in corresponding to the second largest number of requested keywords as key plug-ins; When D max / D min <η, D max The corresponding preset plug-in is determined as the alternative plug-in, where D max is the value corresponding to the maximum number of requested keywords, D min is the value corresponding to the number of the second most requested keywords, and n is the preset ratio threshold; When D max / D min ≥η, a weighted sum is calculated based on the preset score weight, quantity weight, the preset interface stability score of each preset plug-in corresponding to the request keyword, and the number of request keywords corresponding to each preset plug-in to obtain the target score corresponding to each preset plug-in, and the preset plug-in corresponding to the maximum target score is selected as an alternative plug-in.

4. The plug-in calling method based on a large model according to claim 1, characterized in that: The input parameter description text corresponding to the candidate plug-in refers to a text composed of an input parameter structure description text and a parameter format description text extracted from preset parameter information required by the candidate plug-in.

5. The plug-in calling method based on a large model according to claim 4, characterized in that: The target parameter information is converted into a preset parameter format of the corresponding target plug-in and then input into the corresponding target plug-in to obtain the plug-in feedback result, including the following steps: Verify the parameter format of the target parameter information after it is converted into the preset parameter format according to the parameter format description text corresponding to the target plug-in; When the verification is consistent, the converted parameters are input into the corresponding target plug-in through the corresponding API interface to obtain the output result of the target plug-in; When the verification is inconsistent, the obtained parameter format is corrected according to the parameter format description text corresponding to the target plug-in to obtain new parameters, and the new parameters are input into the corresponding target plug-in through the corresponding API interface to obtain the output result of the target plug-in; The output results of all target plug-ins are combined and processed to obtain the plug-in feedback results.

6. The plug-in calling method based on a large model according to claim 1, characterized in that: The method further comprises the steps of: Calculate the text similarity between the received target request text and each historical request text in the historical plug-in selection library; When the maximum text similarity exceeds a preset similarity threshold, the historical plug-in selection result corresponding to the maximum text similarity is sent to the given large language model, so that the given large language model determines the plug-in corresponding to the historical plug-in selection result as the target plug-in.

7. The plug-in calling method based on a large model according to claim 1, characterized in that: The method further comprises the steps of: When the plug-in feedback result is judged to be not in compliance with the preset requirements, a training data set corresponding to each preset plug-in is constructed and input into a given large language model to obtain a model output result; the training data set refers to a data set obtained after expanding the sample input and adding prompt words; According to the output results of the model and the actual results in the training data set, the given large language model is adjusted to obtain a trained target large language model; The target large language model is used as a new given large language model, and the execution step is returned to perform semantic analysis on the target request text and the plug-in description text and input parameter description text corresponding to each candidate plug-in through the given large language model, and according to the semantic analysis result, the target plug-in is screened out from a number of candidate plug-ins until the plug-in feedback result is judged to meet the preset requirements.

8. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the large model-based plug-in calling method as described in any one of claims 1-7.

9. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 8.

Images