Model prompt processing method and device, computer equipment and storage medium

Abstract

The invention relates to a model prompt processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: in response to a model prompt input operation triggered in a model prompt input area, displaying a model prompt input based on the model prompt input operation; in response to a model input data configuration operation triggered in the model input data configuration area, displaying a target input data identifier configured based on the model input data configuration operation; in response to a large model configuration operation triggered in the large model configuration area, displaying a target model identifier configured based on the large model configuration operation; and in response to the large model running operation, displaying a processing result of processing model input data corresponding to the target input data identifier by a large model corresponding to the target model identifier according to the guidance of the model prompt. By adopting the method, the model prompt processing threshold can be reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a model prompt processing method, device, computer equipment and storage medium. Background Art

[0002] With the development of science and technology, big models have emerged. Big models are a type of model in the field of machine learning and artificial intelligence, which are characterized by a large number of parameters and complex structures. Model prompts can be written and input into the big model so that the big model outputs corresponding results based on the model prompts. Model prompts, also known as prompts, refer to an input of a given machine learning model, which is used to guide the machine learning model to generate corresponding outputs. For example, a model prompt can be "The characteristic of a marketing draft is that there are many compliments. Please determine whether this article <article> is a marketing draft." After the model prompt and the article to be analyzed are input into the big model, the big model can determine whether the input article is a marketing draft based on the input model prompt.

[0003] At present, it is mainly business personnel who put forward the writing requirements of model prompts based on specific business, and professional technicians write model prompts based on the writing requirements put forward by business personnel, and return the written model prompts to business personnel. Although professional technicians have a deep understanding of the field of big models, they often lack knowledge reserves in specific business aspects, so there is a cognitive gap in the writing requirements of model prompts, resulting in inaccurate model prompts, and the content output by the big model based on the model prompts is difficult to restore the true intention of the business. Therefore, there is an urgent need for a model prompt writing method that can be oriented to business personnel. Summary of the invention

[0004] Based on this, it is necessary to provide a model prompt processing method, apparatus, computer equipment, computer-readable storage medium and computer program product that can lower the threshold for model prompt processing so that non-professional technicians can also write model prompts and determine the effects of the written model prompts in response to the above-mentioned technical problems.

[0005] In a first aspect, the present application provides a model prompt processing method, the method comprising:

[0006] Displaying a model prompt processing page, and displaying a model prompt input area, a model input data configuration area, and a large model configuration area in the model prompt processing page;

[0007] In response to a model prompt input operation triggered in the model prompt input area, displaying a model prompt input based on the model prompt input operation;

[0008] In response to a model input data configuration operation triggered in the model input data configuration area, displaying a target input data identifier configured based on the model input data configuration operation;

[0009] In response to a large model configuration operation triggered in the large model configuration area, displaying a target model identifier configured based on the large model configuration operation;

[0010] In response to the large model running operation, the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt is displayed.

[0011] In a second aspect, the present application also provides a model prompt processing device, the device comprising:

[0012] A page display module, used to display a model prompt processing page, and display a model prompt input area, a model input data configuration area and a large model configuration area in the model prompt processing page;

[0013] an editing configuration module, for, in response to a model prompt input operation triggered in the model prompt input area, displaying a model prompt input based on the model prompt input operation; in response to a model input data configuration operation triggered in the model input data configuration area, displaying a target input data identifier configured based on the model input data configuration operation; in response to a large model configuration operation triggered in the large model configuration area, displaying a target model identifier configured based on the large model configuration operation;

[0014] A result display module is used to respond to the large model running operation and display the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt.

[0015] In one of the embodiments, the model prompt processing device also includes an adjustment module for, when it is determined based on the processing result that the model prompt meets the tuning conditions, in response to a tuning operation on the model prompt, redisplaying the model prompt processing page and redisplaying the model prompt in the redisplayed model prompt processing page; in response to a modification operation on the redisplayed model prompt, displaying the modified model prompt; using the modified model prompt as a new model prompt, returning to the step of responding to the large model running operation to continue execution.

[0016] In one of the embodiments, the page display module is also used to display a login page of the model prompt self-service processing system, and in response to a login operation triggered in the login page, when the account role of the login account input in the login operation is the target role, display at least one business scenario and a model prompt new creation control; in response to a first selection operation for the at least one business scenario, distinguishably display the target business scenario selected by the first selection operation; in response to a triggering operation for the model prompt new creation control, display the model prompt processing page under the target business scenario.

[0017] In one of the embodiments, the page display module is also used to respond to a trigger operation of creating a new control for the model prompt, and to display a model prompt processing page for the target business scenario when the login account has the authority to create a model prompt for the target business scenario.

[0018] In one of the embodiments, the target input data is identified as a data link of the model input data; the model input data configuration area includes a first data input box; and the editing configuration module is also used to respond to a link input operation triggered in the first data input box, and display in the first data input box a data link of the model input data input based on the link input operation.

[0019] In one embodiment, the model input data configuration area includes a second data input box and a model input data selection control; the editing configuration module is also used to display a data list in response to a trigger operation on the model input data selection control; in response to a second selection operation on multiple input data identifiers in the data list, a target input data identifier selected by the second selection operation from the multiple input data identifiers is displayed in the second data input box.

[0020] In one embodiment, the large model configuration area includes a model identification display box; the model identification display box includes a drop-down control; the editing configuration module is also used to display multiple first candidate model identifications in response to a trigger operation on the drop-down control; in response to a third selection operation on the multiple first candidate model identifications, display a target model identification selected from the multiple first candidate model identifications based on the third selection operation.

[0021] In one of the embodiments, the editing configuration module is also used to display a model prompt template list, which includes at least one template identifier and an operation control sequence corresponding to each template identifier; in response to a triggering operation on an editing control in the operation control sequence, determining a triggered template identifier corresponding to the triggered editing control; displaying a triggered model prompt template corresponding to the triggered template identifier; and obtaining a model prompt in response to a modification operation on the triggered model prompt template.

[0022] In one of the embodiments, the result display module is also used to respond to the large model running operation, input the model prompt and the model input data corresponding to the target input data identifier into the large model corresponding to the target model identifier; call the large model corresponding to the target model identifier so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier based on the model prompt, obtains the processing result, and displays the processing result.

[0023] In one of the embodiments, the result display module is also used to display a processing result display page, and display the processing result and the model prompt in the processing result display page; based on the correlation between the processing result and each sentence in the model prompt, the sentences in the model prompt in the processing result display page are differentiated and displayed; model prompt optimization information is generated based on the differentiated and displayed sentences, and the model prompt optimization information is displayed.

[0024] In one of the embodiments, a save control is displayed in the model prompt processing page; the model prompt processing device also includes a save module, which is used to generate a target model prompt template and a target template identifier corresponding to the target model prompt template based on the model prompt in response to a trigger operation on the save control; in the model prompt template list under the business scenario corresponding to the model prompt, the target template identifier and the operation control sequence corresponding to the target template identifier are displayed.

[0025] In one of the embodiments, the operation control sequence corresponding to the target template identifier also includes a detail control; the model prompt processing device also includes a multi-model comparison module, which is used to display a multi-model comparison page in response to a trigger operation on the detail control, and display a multi-model comparison control in the multi-model comparison page; in response to a trigger operation on the multi-model comparison control, multiple second candidate model identifiers are displayed; in response to a first multiple-selection operation on the multiple second candidate model identifiers, multiple comparison model identifiers selected from the multiple second candidate model identifiers based on the first multiple-selection operation are displayed; in response to a multi-model running operation, a multi-model processing result is displayed; the multi-model processing result is the result of processing the model input data corresponding to the target input data identifier by the large model corresponding to each of the multiple comparison model identifiers based on the model prompt.

[0026] In one of the embodiments, the multi-model processing result includes the large model output results corresponding to each of the multiple comparison model identifiers; the multi-model comparison module is also used to display the output results of each large model in the multi-model processing result on the same screen.

[0027] In one of the embodiments, the model prompt processing device also includes a batch running module, which is used to display a batch processing page in response to a batch processing operation on the model prompt, and display a multi-model selection area and a data batch input area in the batch processing page; display a plurality of third candidate model identifiers in response to a trigger operation on the multi-model selection area; display a plurality of batch processing model identifiers selected from the plurality of third candidate model identifiers through the second multiple selection operation in response to a second multiple selection operation on the plurality of third candidate model identifiers; display a target data set identifier input through the model input data batch input operation in response to a model input data batch input operation triggered in the data batch input area; and display a batch processing result of batch processing of the model input data set corresponding to the target data set identifier by the large model corresponding to each of the plurality of batch processing model identifiers based on the model prompt in response to the batch processing operation.

[0028] In one of the embodiments, the batch processing result includes a result subset corresponding to each model input data in the model input data set corresponding to the target data set identifier; each of the result subsets includes the result of the large model output corresponding to each of the batch processing model identifiers;

[0029] The batch running module is also used to display each model input data in the model input data set corresponding to the target data set identifier; for each model input data in the model input data set corresponding to the target data set identifier, a result subset corresponding to the model input data is displayed in a blank area corresponding to the model input data.

[0030] In one of the embodiments, the model prompt processing device also includes a publishing application module, which is used to display a model binding control in response to a publishing operation on the model prompt; display multiple fourth candidate model identifiers in response to a trigger operation on the model binding control; in response to a fourth selection operation on the multiple fourth candidate model identifiers, bind the binding model identifier selected by the fourth selection operation from the multiple fourth candidate model identifiers with the model prompt to obtain a binding result, and generate a publishing request based on the binding result; send the publishing request to a terminal corresponding to a business approver; the sent publishing request is used to trigger the business approver to perform a publishing approval based on the model prompt and the large model bound to the model prompt, and publish the model prompt if the approval is passed.

[0031] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps in any one of the model prompt processing methods provided in the embodiments of the present application are implemented.

[0032] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any model prompt processing method provided in the embodiments of the present application are implemented.

[0033] In a fifth aspect, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps in any model prompt processing method provided in the embodiments of the present application.

[0034] The above-mentioned model prompt processing method, device, computer equipment, storage medium and computer program product can display the model prompt input area, model input data configuration area and large model configuration area based on the model prompt processing page by displaying the model prompt processing page. By displaying the model prompt input area, the model prompt can be written based on the model prompt input area. By displaying the model input data configuration area, the model input data can be configured based on the model input data configuration area. By displaying the large model configuration area, the large model can be configured based on the large model configuration area. By writing model prompts, configuring model input data and large models, the configured large model can be called to process the configured model input data based on the written model prompts, and obtain and display the processing results. Since only simple input and configuration operations are required to realize the editing of model prompts and to trigger the operation of large models through model prompts, and obtain the processing results of large model outputs, even non-professional technicians can easily write and use model prompts, which reduces the threshold for writing and using model prompts and improves the efficiency of writing and using model prompts.

[0035] In addition, compared with the traditional method where users propose business requirements to professional technicians based on their own needs, professional technicians write model prompts based on the business requirements proposed by users. Since users can directly participate in the writing of model prompts, it can reduce the probability that the written model prompts are difficult to restore the true business intentions due to the deviation in professional technicians' understanding of the needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 An application environment diagram of a model prompt processing method in an embodiment;

[0037] Figure 2 A schematic diagram of a flow chart of a model prompt processing method in one embodiment;

[0038] Figure 3 A schematic diagram of a model prompt processing page in one embodiment;

[0039] Figure 4 A schematic diagram of fine-tuning in one embodiment;

[0040] Figure 5 A schematic diagram of a workflow page in one embodiment;

[0041] Figure 6 is a schematic diagram of a first data input box in one embodiment;

[0042] Figure 7 A schematic diagram of the configuration of model input data in one embodiment;

[0043] Figure 8A schematic diagram of a model prompt template list in one embodiment;

[0044] Fig. 9 A schematic diagram of a model prompt display in one embodiment;

[0045] Fig.10 A schematic diagram of a multi-model comparison page in an embodiment;

[0046] Fig.11 A schematic diagram of displaying multiple processing results on the same screen in one embodiment;

[0047] Fig.12 A schematic diagram of batch processing pages in one embodiment;

[0048] Fig.13 A schematic diagram of the overall process of creating and publishing a model prompt in one embodiment;

[0049] Fig.14 An interactive schematic diagram of a model prompt for self-service creation and publishing in one embodiment;

[0050] Fig.15 A schematic diagram of the architecture of a model prompt self-service processing system in one embodiment;

[0051] Fig.16 A schematic diagram of the overall process written for a model prompt in one embodiment;

[0052] Fig.17 A schematic diagram of the overall process of adjusting a model prompt in an embodiment;

[0053] Fig.18 A schematic diagram of model prompt release in one embodiment;

[0054] Fig.19 It is a structural block diagram of a model prompt processing device in one embodiment;

[0055] Fig. 20 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] The model prompt processing method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other servers. Take the terminal 102 and the server 104 as an example to be used in conjunction to execute the model prompt processing method provided in the embodiment of the present application. The user can trigger the terminal 102 to display the model prompt processing page, and edit the model prompt through the model prompt processing page, and configure the model input data and the large model through the model prompt processing page to obtain configuration information. The terminal 102 can send the configuration information and the edited model prompt to the server 104, so that the server 104 determines the corresponding large model and model input data based on the configuration information, inputs the model input data and the model prompt into the large model, and obtains the processing result of the large model output. The server 104 returns the processing result to the terminal 102, and then the terminal 102 can display the processing result. The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0058] It should be noted that the words "first", "second" and similar terms used in this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Unless the context clearly indicates otherwise, the singular form "one", "an" or "the" and similar terms do not indicate a quantity restriction, but rather indicate the presence of at least one. The quantities of "multiple" or "multiple copies" mentioned in the various embodiments of the present application all refer to the quantity of "at least two", for example, "multiple" refers to "at least two", and "multiple copies" refers to "at least two copies".

[0059] In one embodiment, Figure 2 As shown, a model prompt processing method is provided, which is applied to Figure 1 The terminal in is used as an example to illustrate, including the following steps:

[0060] Step 202: Display the model prompt processing page, and display the model prompt input area, the model input data configuration area and the large model configuration area in the model prompt processing page.

[0061] Specifically, when a user needs to edit a model prompt, the user can trigger the terminal to display the model prompt processing page. The model prompt processing page may include a model prompt input area, a model input data configuration area, and a large model configuration area. For example, Figure 3 , Figure 3 A schematic diagram of a model prompt processing page in one embodiment is shown. Figure 3 301 is a model prompt input area, which is used to receive the model prompt input by the user; Figure 3 302 is a model input data configuration area, which is used to configure the model input data; Figure 3 303 is the large model configuration area, which is used to configure the large model.

[0062] In one embodiment, a model prompt self-service processing system is deployed in the terminal, and the user can start the model prompt self-service processing system, and then the terminal displays the home page of the model prompt self-service processing system. A navigation bar can be displayed in the home page, and the user can use the navigation bar to trigger the terminal to display the model prompt processing page provided by the model prompt self-service processing system. For example, the user can click the "Work" button control in the navigation bar to make the terminal display the model prompt processing page.

[0063] Step 204 , in response to a model prompt input operation triggered in the model prompt input area, displaying a model prompt based on the model prompt input operation input.

[0064] Specifically, the user can edit the model prompt in the model prompt input box in the model prompt input area, so that the terminal can display the model prompt input by the user in the model prompt input box. Among them, the model prompt is also called prompt, which refers to an input of a given machine learning model, which is used to guide the machine learning model to generate the corresponding output. The model prompt can be a question, a sentence, a paragraph of text, or a set of instructions, the purpose of which is to guide the machine learning model to output the desired result.

[0065] In one embodiment, reference Figure 3 , Figure 3 A model prompt input box may be displayed in the model prompt input area 301, and the user may directly input a model prompt in the model prompt input box according to his / her own needs. For example, when the user is a business operator and has a business need for article repetitiveness judgment, the business operator may write a model prompt for guiding the large model to perform article repetitiveness judgment in the model prompt input box, and then the written model prompt may be displayed in the model prompt input box.

[0066] Step 206 , in response to the model input data configuration operation triggered in the model input data configuration area, display the target input data identifier configured based on the model input data configuration operation.

[0067] Specifically, in addition to editing the model prompts, users can also configure the model input data. Model input data refers to the data input into the big model, so that the big model can process the model input data accordingly according to the guidance of the model prompts. For example, in the above example, when the user has a business need to judge the repetitiveness of articles, and after editing the model prompts, the user can also configure the articles used for repetitive judgment. The articles used for repetitive judgment are the configured model input data.

[0068] In one embodiment, when the model input data needs to be configured, the user can trigger the model input data configuration operation in the model prompt input area, and then the terminal can respond to the model input data configuration operation and display the input data identification input based on the model input data configuration operation. The input data identification refers to the identification of the model input data. For example, refer to Figure 3 , an input box may be displayed in the model input data configuration area 302, in which the user may enter a data identifier, which is the target input data identifier configured based on the model input data configuration operation. For example, the target input data identifier may be a link, and in the above example, the business personnel may enter a link of an article for repeatability judgment in the input box.

[0069] In one of the embodiments, an input box may be displayed in the model prompt input area, and the user may directly input the model input data into the input box. In order to make the interface more concise, after the user inputs the model input data, the terminal may display the target input data identifier of the input model input data in the input box.

[0070] Step 208 , in response to the large model configuration operation triggered in the large model configuration area, displaying the target model identifier configured based on the large model configuration operation.

[0071] Specifically, users can not only edit model prompts and configure model input data, but also configure large models. For example, users can specify which large model to use to process model prompts and model input data. Among them, large models refer to "large parameter" models trained using large-scale data and powerful computing power. These models usually have high versatility and generalization capabilities and can be applied to natural language processing, image recognition, speech recognition and other fields. Large models can be divided into large language models, large visual models, multimodal large models, basic large models, and so on.

[0072] In one of the embodiments, the user can trigger the large model configuration operation through the large model configuration area, and then the terminal can respond to the large model configuration operation and display the target model identifier configured based on the large model configuration operation. For example, the large model configuration area can display the model identifiers of multiple large models, so that the user can select one from the model identifiers of multiple large models as the target model identifier, and then the terminal can display the target model identifier selected by the user. Furthermore, in addition to displaying the model identifiers of multiple large models, the large model configuration area can also display the model introduction corresponding to each model identifier, and the user can select the target model identifier based on the displayed model introduction.

[0073] Step 210, in response to the large model running operation, displaying the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt.

[0074] Specifically, after the model prompt is edited and the model input data and the big model are configured, the user can trigger the big model running operation, and then the terminal can respond to the big model running operation, obtain the big model corresponding to the target model identifier, and according to the guidance of the model prompt, process the model input data corresponding to the target input data identifier, and display the processing result.

[0075] In one of the embodiments, a control for triggering the operation of a large model may be displayed in the model prompt processing page. For example, a "Go to Run" button control may be displayed in the model prompt processing page. When the user clicks the "Go to Run" button control, it can be considered that the user has triggered the large model operation. At this time, the terminal can input the model prompt and the model input data corresponding to the target input data identifier into the large model corresponding to the target model identifier, so that the large model outputs the processing result.

[0076] In one embodiment, a "preview" button control may also be displayed in the model prompt output area. When the user clicks the "preview" button control, the terminal may display a preview interface, and display the model prompt edited by the user in the preview interface, and display a "debug" button control. When the user clicks the "debug" button control, it may also be considered that the user has triggered the large model running operation.

[0077] In one embodiment, a "Save" button control may be displayed in the model prompt processing page. When the user clicks the "Save" button control, the terminal may store the model prompt edited by the user, the configured input data identifier and the large model identifier as the target model prompt template. The user may trigger the terminal to display the model prompt template list, and trigger the large model running operation through the running operation control corresponding to the target model prompt template in the model prompt template list. For example, the user may click the "Run" button control corresponding to the target model prompt template to trigger the large model running operation.

[0078] In the above-mentioned model prompt processing method, by displaying the model prompt processing page, the model prompt input area, the model input data configuration area and the large model configuration area can be displayed based on the model prompt processing page. By displaying the model prompt input area, the model prompt can be written based on the model prompt input area. By displaying the model input data configuration area, the model input data can be configured based on the model input data configuration area. By displaying the large model configuration area, the large model can be configured based on the large model configuration area. By writing model prompts, configuring model input data and large models, the configured large model can be called to process the configured model input data based on the written model prompts, and obtain and display the processing results. Since only simple input and configuration operations are required to realize the editing of model prompts and to trigger the operation of large models through model prompts, and obtain the processing results of large model outputs, even non-professional technicians can easily write and use model prompts, which reduces the threshold for writing and using model prompts and improves the efficiency of writing and using model prompts.

[0079] In addition, compared with the traditional method where users propose business requirements to professional technicians based on their own needs, professional technicians write model prompts based on the business requirements proposed by users. Since users can directly participate in the writing of model prompts, it can reduce the probability that the written model prompts are difficult to restore the true business intentions due to the deviation in professional technicians' understanding of the needs.

[0080] In one of the embodiments, the method further includes: in a case where it is determined based on the processing results that the model prompt meets the tuning conditions, in response to a tuning operation on the model prompt, redisplaying the model prompt processing page, and redisplaying the model prompt in the redisplayed model prompt processing page; in response to a modification operation on the redisplayed model prompt, displaying the modified model prompt; using the modified model prompt as a new model prompt, and returning to the step of responding to the large model running operation to continue execution.

[0081] Specifically, the user can determine whether the written model prompt needs to be tuned based on the displayed processing results. For example, when the processing results output by the large model are incorrect, it can be considered that the model prompt has not been able to correctly guide the large model to output the expected processing results. At this time, the model prompt needs to be adjusted to meet the tuning conditions. When the model prompt meets the tuning conditions, the user can trigger a tuning operation for the model prompt, and then the terminal can respond to the tuning operation and re-display the model prompt processing page. For example, the terminal can jump from the processing result display page to the model prompt processing page, and re-display the model prompt in the re-displayed model prompt processing page. Furthermore, the user can modify the re-displayed model prompt, and the terminal can use the modified model prompt as a new model prompt and return to the step of responding to the large model running operation to continue execution. Iterate in this way until the model prompt does not meet the tuning conditions.

[0082] For example, in addition to the model prompt, the re-displayed model prompt processing page may also display the target input data identifier and the target model identifier. The user can modify the displayed model prompt to obtain the modified model prompt. The user can trigger the large model running operation so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier according to the guidance of the modified model prompt to obtain the processing result. Then, the terminal can display the processing result output by the large model based on the modified model prompt. The user can continue to judge whether the modified model prompt needs to be tuned again based on the displayed processing result. If it needs to be tuned again, perform the above steps again until the model prompt does not meet the tuning conditions and can guide the large model to output the expected results.

[0083] In one embodiment, the model prompts can be tuned by the idea of ​​SFT (Supervised Fine-Tuning). Figure 4 SFT means pre-training a neural network model on the source data set, namely the source model. Then a new neural network model is created, namely the target model. The target model copies all the model designs and parameters of the source model except the output layer. When fine-tuning, the model parameters of the initial output layer are randomly set. When training the target model, the output layer will be trained from scratch, and the parameters of the remaining layers are fine-tuned based on the parameters of the source model. When it is necessary to tune the model prompt, the model prompt can be processed according to the SFT idea. For example, when the model prompt includes the sentence "If more than 50% of a promotional article focuses on the above three points, it is considered to be a heavy promotional article", the parameter "50%" in this sentence can be fine-tuned, for example, "50%" can be fine-tuned to "60%", and the rest of the content in the model prompt remains unchanged. Figure 4 A schematic diagram of fine-tuning in one embodiment is shown.

[0084] In one embodiment, the terminal may pre-evaluate the part of the model prompt that can be fine-tuned, and highlight the part of the model prompt that can be fine-tuned. For example, the terminal may use the numerical value in the model prompt as the part that can be fine-tuned, and highlight the numerical value in the model prompt.

[0085] In one embodiment, the model prompt can also be tuned by the idea of ​​DPO (Direct Preference Optimization). DPO does not need to explicitly define the reward function, but selects the result that is more in line with human preferences as the training target by comparing the results output by different models. In practical applications, DPO can train the model by constructing a sorted data set. For example, for a text generation task, a sorted data set containing good text and bad text can be constructed, and then the model can be trained to generate text that is more in line with human preferences. When the model prompt is tuned based on the DOP idea, the processing result based on the output of the currently adjusted model prompt can be compared with the processing result based on the output of the previously adjusted model prompt to determine which processing result is more correct, and the model prompt corresponding to the more correct processing result is used as the better model prompt, and the other model prompt is used as the worse model prompt. By comparing the difference between the better model prompt and the worse model prompt, the features of the better model prompt can be learned, and the model prompt can be tuned based on the learned features.

[0086] In one of the embodiments, after the user has written the model prompt and configured the target input data identifier and the target model identifier, the user can click the "Save" button control, and the terminal can save the written model prompt, the configured target input data identifier and the target model identifier as a target model prompt template. When the model prompt meets the tuning conditions, the user can trigger the terminal to display a model prompt template list including multiple template identifiers, and click the template identifier of the target model prompt template in the model prompt template list. In response to the user's click operation on the template identifier, the terminal can redisplay the model prompt processing page, and redisplay the model prompt, the target input data identifier and the target model identifier in the redisplayed model prompt processing page.

[0087] In the above embodiment, by displaying the processing results, the user can determine whether the model prompt needs to be tuned based on the displayed processing results. When it is determined that the model prompt needs to be tuned, the user can conveniently tune the model prompt based on the re-displayed model prompt processing page, thereby improving the efficiency of model prompt tuning. In addition, since the above process can be repeated, the model prompt can be continuously adjusted until the model prompt can guide the large model to output the expected results. This improves the accuracy of the model prompt finally obtained.

[0088] In one of the embodiments, the terminal can display on the same screen the processing results output by the large model after each optimization of the model prompt, so that the user can determine the next adjustment direction for the model prompt based on the processing results displayed on the same screen.

[0089] In one of the embodiments, a model prompt processing page is displayed, including: displaying a login page of a model prompt self-service processing system, and in response to a login operation triggered in the login page, when the account role of the login account input in the login operation is a target role, displaying at least one business scenario and a model prompt new creation control; in response to a first selection operation for at least one business scenario, distinguishingly displaying the target business scenario selected by the first selection operation; in response to a triggering operation for a model prompt new creation control, displaying a model prompt processing page under the target business scenario.

[0090] Specifically, the model prompt processing page can be provided by the model prompt self-service processing system. Before displaying the model prompt processing page, the user can first log in to the model prompt self-service processing system. For example, the user can start the model prompt self-service processing system through the terminal, and then the terminal can display the login page of the model prompt self-service processing system. The user can enter the login account and login password through the login page. When the login is successful based on the login account and login password, and the account role of the login account is determined to be the target role based on the login account input in the login operation, the terminal can display at least one business scenario and a new model prompt control. For example, refer to Figure 5 The terminal may display a workflow page, and display at least one business scenario 501 in the workflow page, and display a model prompt new control 502. Business scenarios may be, for example, information scenarios, search scenarios, etc.; information scenarios may include information article quality scenarios, information AI summary scenarios, etc.; search scenarios may include intelligent search scenarios, etc.

[0091] Furthermore, the user may select one from at least one of the displayed business scenarios as a target business scenario, and then the terminal may highlight the target business scenario in response to the first selection operation for at least one business scenario. When the user triggers the model prompt creation control, the terminal may respond to the triggering operation for the model prompt creation control and display the model prompt processing page for the target business scenario. For example, the model prompt creation control may specifically be a "model prompt creation" button control. When the user clicks the "model prompt creation" button control, the terminal may display the model prompt processing page for the target business scenario. The model prompt processing page associated with the target business scenario refers to a page used to create a model prompt for the target business scenario. For example, the model prompt processing page associated with information refers to a page used to create a model prompt for use in an information scenario. And when the user clicks the "Save" button control in the model prompt processing page, the generated model prompt template will be stored in the model prompt template library associated with the target business scenario. Figure 5 A schematic diagram of a workflow page in an embodiment is shown.

[0092] In one of the embodiments, different account roles have different functions. Account roles can be divided into system administrator role, business approver role, business operator role and visitor. The system administrator role has all permissions at the system development level, as well as category setting permissions (including business, scenario, and module); the business approver is responsible for approving the release and launch application of the business operator; the business operator role has the permissions to create, modify, debug, delete, and publish model prompts, as well as the permissions to create, manage, and evaluate model input data sets; the visitor has browsing-only permissions. When the account role of the login account input by the login operation is the business operator role, the account role of the login account input by the login operation is determined to be the target role.

[0093] In the above embodiment, when it is determined that the account role of the login account input in the login operation is the target role, the model prompt creation control is displayed, which can improve the security of model prompt creation. In addition, by selecting the business scenario, the model prompt under the selected target business scenario can be created.

[0094] In one of the embodiments, in response to a trigger operation for creating a new control for a model prompt, a model prompt processing page for a target business scenario is displayed, including: in response to a trigger operation for creating a new control for a model prompt, when the logged-in account has permission to create a model prompt for the target business scenario, a model prompt processing page for the target business scenario is displayed.

[0095] Specifically, users of the target role can be divided into business lines according to business scenarios. For example, users of the target role can be divided into editors, customer service personnel, IT personnel, and so on. Users under different business lines correspond to different business scenarios. For example, editors correspond to information scenarios, customer service personnel correspond to search scenarios, and so on. Users corresponding to the same business scenario can view, create, modify, and run the model prompts and model input data sets under the business scenario. Users across business scenarios can only browse and cannot perform other operations. When a user triggers a new model prompt control, the terminal can respond to the trigger operation of the new model prompt control to determine whether the user has the authority to create a model prompt under the target business scenario. If so, the model prompt processing page under the target business scenario is displayed. If not, a prompt message can be displayed to prompt the user that he does not have the authority to create a model prompt.

[0096] In this embodiment, by making users under different business lines correspond to different business scenarios, user permissions can be effectively isolated based on business scenarios, so that risk prevention can be carried out based on effective isolation, thereby improving the security of model prompt processing.

[0097] In one of the embodiments, the target input data identifier is a data link of the model input data; the model input data configuration area includes a first data input box; in response to a model input data configuration operation triggered in the model input data configuration area, the target input data identifier configured based on the model input data configuration operation is displayed, including: in response to a link input operation triggered in the first data input box, a data link of the model input data input based on the link input operation is displayed in the first data input box.

[0098] Specifically, the target input data identifier may be a data link of the model input data. A first data input box may be displayed in the model input data configuration area, for example, Figure 6 The first data input box 601 is shown. The user can input a link in the first data input box, so that the terminal can respond to the link input operation triggered in the first data input box and display the data link input by the link input operation in the first data input box. Then the subsequent large model can obtain the model input data based on the data link. Figure 6 A schematic diagram of a first data input box in one embodiment is shown. In one embodiment, the data link can be a SNP link, an external network link, etc.

[0099] In this embodiment, the model input data part, by opening up the business backend management platform, allows users to only enter the data link in the first data input box, without having to enter the complete model input data into the first data input box, thereby reducing manual entry costs and improving the efficiency of obtaining model input data.

[0100] In one of the embodiments, the model input data configuration area includes a second data input box and a model input data selection control; in response to a model input data configuration operation triggered in the model input data configuration area, a target input data identifier configured based on the model input data configuration operation is displayed, including: in response to a trigger operation on the model input data selection control, a data list is displayed; in response to a second selection operation on multiple input data identifiers in the data list, a target input data identifier selected from the multiple input data identifiers by the second selection operation is displayed in the second data input box.

[0101] Specifically, the model input data configuration area may include a second data input box and a model input data selection control. Figure 7 , may include a second data input box 701 and a model input data selection control 702. When the user triggers the model input data selection control, for example, when the user clicks the model input data selection control, the terminal may display a data list, and the user may select one from multiple input data identifiers displayed in the data list, and then the terminal may respond to the user's second selection operation on multiple input data identifiers in the data list, and display the target input data identifier selected by the second selection operation from multiple input data identifiers in the second data input box. It is easy to understand that the first data input box and the second data input box may be the same input box or different input boxes.

[0102] In one embodiment, the terminal may display a data list 703. When the input data is identified as an article title, multiple article titles and a selection control corresponding to each article title may be displayed in the data list 703. The user may select an article title by selecting the control, and the terminal then displays the article title selected by the user in the second data input box. Figure 7 A schematic diagram of the configuration of model input data in an embodiment is shown.

[0103] In the above embodiment, by pre-setting multiple model input data, the user can directly select the required model input data from the pre-set multiple model input data without having to search for available model input data from massive Internet resources, which improves the efficiency of obtaining model input data.

[0104] In one of the embodiments, the large model configuration area includes a model identification display box; the model identification display box includes a drop-down control; in response to a large model configuration operation triggered in the large model configuration area, a target model identification configured based on the large model configuration operation is displayed, including: in response to a trigger operation on the drop-down control, a plurality of first candidate model identifications are displayed; in response to a third selection operation on the plurality of first candidate model identifications, a target model identification selected from the plurality of first candidate model identifications based on the third selection operation is displayed.

[0105] Specifically, a model identification display box may be displayed in the large model configuration area. A drop-down control may be displayed in the model identification display box. When the user triggers the drop-down control, the terminal may display multiple first candidate model identifications in response to the user's triggering operation on the drop-down control. The user may select one from the displayed multiple first candidate model identifications, and then the terminal may respond to the user's third selection operation on the multiple first candidate model identifications, use the first candidate model identification selected by the third selection operation as the target model identification, and display the target model identification in the model identification display box.

[0106] In this embodiment, the user only needs to perform a simple selection operation to configure the large model, thereby improving the configuration efficiency of the large model configuration.

[0107] In one of the embodiments, the method further includes: displaying a model prompt template list, the model prompt template list including at least one template identifier and an operation control sequence corresponding to each template identifier; in response to a triggering operation on an editing control in the operation control sequence, determining a triggered template identifier corresponding to the triggered editing control; displaying a triggered model prompt template corresponding to the triggered template identifier; and in response to a modification operation on the triggered model prompt template, obtaining a model prompt.

[0108] Specifically, in addition to creating a new model prompt by creating a new control through a model prompt, a model prompt can also be imported from an existing template. For example, the terminal can display a model prompt template list, and the model prompt template list includes at least one template identifier and an operation control sequence corresponding to each template identifier. The operation control sequence includes a run control, a view control, a batch run control, an edit control, and a delete control. The user performs a trigger operation on the edit operation control in the operation control sequence, and then the terminal can determine the template identifier corresponding to the triggered edit control in response to the trigger operation on the edit control. The template identifier corresponding to the triggered edit control is called the triggered template identifier, and the model prompt template corresponding to the triggered template identifier is called the triggered model prompt template. The terminal can display the triggered model prompt template corresponding to the triggered template identifier, and then the user can directly modify the model prompt in the triggered model prompt template to obtain the model prompt. In addition to the model prompt, the displayed triggered model prompt template may also display the input data identifier and the model identifier. In addition to modifying the model prompt, the user may also modify the input data identifier and the model identifier in the triggered model prompt template.

[0109] Exemplarily, when the terminal displays at least one business scenario, after the user selects the target business scenario from at least one business scenario, the terminal may display at least one subcategory under the target business scenario, and the user may select the target subcategory from at least one subcategory, and then the terminal displays the model prompt template list under the target subcategory. Among them, the template identifier in the model prompt template list under the target subcategory may be consistent with the name of the target subcategory. For example, referring to 8, when the user selects the information article quality scenario, the terminal may display at least one subcategory under the information article quality scenario, and at least one subcategory may include AI judgment of repetitive articles, AI judgment of regional articles, and AI filtering of soft articles and marketing drafts. The user may select one from at least one subcategory, for example, the user may select AI filtering of soft articles and marketing drafts from at least one subcategory, so that the terminal displays the model prompt template list 801 under AI filtering of soft articles and marketing drafts, and when the user clicks the edit control in the operation control sequence 802, the terminal may display the triggered model prompt template corresponding to the clicked edit control. Figure 8 A schematic diagram of a model prompt template list in an embodiment is shown.

[0110] In this embodiment, the user can directly modify the model prompt template to obtain a model prompt that meets his or her needs. Compared with re-editing a model prompt, the editing efficiency of the model prompt can be improved.

[0111] In one of the embodiments, in response to a large model running operation, a processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt is displayed, including: in response to the large model running operation, the model prompt and the model input data corresponding to the target input data identifier are input into the large model corresponding to the target model identifier; the large model corresponding to the target model identifier is called so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier based on the model prompt, obtains the processing result, and displays the processing result.

[0112] Specifically, in response to the large model running operation, the terminal inputs the model prompt and the model input data corresponding to the target input data identifier into the large model corresponding to the target model identifier, so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier according to the guidance given by the model prompt, and obtains the processing result of the large model output corresponding to the target model identifier. When the processing result is obtained, the terminal can display the processing result. In one embodiment, the terminal can display the model prompt and the processing result on the same screen. For example, refer to Fig. 9 , Fig. 9 901 in the figure is the model prompt. Fig. 9 902 in the figure is the processing result. Fig. 9 A schematic diagram of model prompt display in one embodiment is shown. By displaying the processing result and the model prompt on the same screen, the user can intuitively understand which model prompt the processing result is based on, thereby improving the user experience.

[0113] In one embodiment, the terminal can obtain the standard processing result corresponding to the model input data, and display the standard processing result, model prompt and processing result on the same screen. Among them, the standard processing result can be a result that the user expects the large model to output. When the user configures the model input data, the standard processing result can also be input at the same time. When the processing result output by the large model is inconsistent with the standard processing result, the terminal can give a prompt that the processing result is inconsistent. At this time, it can be considered that the model prompt written by the user is inaccurate and meets the tuning conditions. The user can tune the model prompt until the processing result output by the large model is consistent with the standard processing result.

[0114] In this embodiment, by displaying the processing results, it is convenient for users to judge whether the model prompt needs to be tuned based on the displayed processing results, thereby improving the user experience. In addition, the user only needs to trigger the large model operation, and the large model corresponding to the configured target model identifier can automatically process the model input data corresponding to the configured target input data identifier based on the written model prompt, and output the processing results, thereby reducing labor costs and improving the efficiency of obtaining processing results.

[0115] In one of the embodiments, displaying the processing results includes: displaying a processing result display page, and displaying the processing results and model prompts on the processing result display page; distinguishing and displaying the sentences in the model prompts on the processing result display page according to the correlation between the processing results and the sentences in the model prompts; generating model prompt optimization information based on the distinguished and displayed sentences, and displaying the model prompt optimization information.

[0116] Specifically, when the terminal obtains the processing result output by the large model, the terminal can display the processing result display page, and display the processing result and model prompt in the processing result display page. While outputting the processing result, the large model can also output the correlation between the processing result and each sentence in the model prompt. The correlation between the processing result and each sentence in the model prompt is determined according to the importance of the sentence in the model prompt to the processing result output by the large model. Compared with sentence A, the large model relies more on sentence B to output the processing result, then the correlation between the processing result and sentence B is greater than the correlation between the processing result and sentence A. The sentences in the model prompt can be distinguished and displayed according to the correlation. For example, sentences with a correlation greater than the first correlation threshold are marked in red, sentences with a correlation less than the first correlation threshold but greater than the second correlation threshold are marked in yellow, and sentences with a correlation less than the second correlation threshold are marked in gray, and so on.

[0117] Furthermore, the terminal may generate model prompt optimization information based on distinguishing the displayed sentences. For example, the terminal may select the sentences marked in red and yellow, and generate model prompt optimization information "it is recommended to adjust and optimize the selected sentences first".

[0118] In this embodiment, by distinguishing and displaying sentences according to relevance, users can distinguish the key sentences in the model prompts based on the distinguished and displayed sentences, and then can prioritize the key sentences for adjustment based on the model prompt optimization information. Compared with randomly adjusting the sentences in the model prompts, prioritizing the key sentences for adjustment can achieve better tuning effects and reduce the number of tuning times.

[0119] In one of the embodiments, a save control is displayed in the model prompt processing page; the method also includes: in response to a trigger operation for the save control, generating a target model prompt template and a target template identifier corresponding to the target model prompt template based on the model prompt; in a model prompt template list under the business scenario corresponding to the model prompt, displaying the target template identifier and an operation control sequence corresponding to the target template identifier.

[0120] Specifically, after the user has written the model prompt and configured the target input data identifier and the target model identifier, the user can click the "Save" button control, and then the terminal can use the written model prompt, the configured model input data and the large model as the target model prompt template, and store the target model prompt template in the model prompt template library under the business scenario corresponding to the model prompt. Further, when displaying the model prompt template list corresponding to the model prompt template library, the target template identifier corresponding to the target model prompt template can be displayed in the model prompt template list, and the operation control sequence corresponding to the target template identifier can be displayed, so that the user can further process the target model prompt template based on the operation control in the operation control sequence, for example, edit it through the editing control in the operation control sequence. When the model prompt in the target model prompt template is modified through the editing control and the "Save" button control is clicked, the model prompt in the target model prompt template is automatically updated to the modified model prompt.

[0121] In one embodiment, the operation control in the operation control sequence may specifically be an operation button control. For example, the edit control in the operation control sequence may specifically be an edit button control.

[0122] In one of the embodiments, the template name of the target model prompt template is used as the target template identifier.

[0123] In this embodiment, by saving the model prompt written by the user and the target input data identifier and the target model identifier as a target model prompt template, subsequent modification of the model prompt, the target input data identifier and the target model identifier can be facilitated.

[0124] In one of the embodiments, the operation control sequence corresponding to the target template identifier also includes a detail control; the above method also includes: in response to a trigger operation on the detail control, displaying a multi-model comparison page, and displaying a multi-model comparison control in the multi-model comparison page; in response to a trigger operation on the multi-model comparison control, displaying multiple second candidate model identifiers; in response to a first multiple-selection operation on multiple second candidate model identifiers, displaying multiple comparison model identifiers selected from multiple second candidate model identifiers based on the first multiple-selection operation; in response to a multi-model running operation, displaying a multi-model processing result; the multi-model processing result is the result of processing the model input data corresponding to the target input data identifier by the large model corresponding to each of the multiple comparison model identifiers based on the model prompt.

[0125] Specifically, the operation control sequence corresponding to the target template identifier may also include a detail control, and the user may trigger the multi-model comparison through the detail control. When the user triggers the detail control, the terminal may display a multi-model comparison page in response to the triggering operation on the detail control, and the multi-model comparison page may display a multi-model comparison control. For example, referring to Fig.10 , Fig.10 A schematic diagram of a multi-model comparison page in an embodiment is shown. Fig.10 1001 in is a multi-model comparison control. When the user triggers the multi-model comparison control, the terminal can display multiple second candidate model identifiers in response to the triggering operation for the multi-model comparison control. The user can select from the multiple second candidate model identifiers displayed, so that the terminal responds to the first multiple selection operation for the multiple second candidate model identifiers and determines the model identifier selected from the multiple second candidate model identifiers based on the first multiple selection operation. The model identifier selected from the multiple second candidate model identifiers based on the first multiple selection operation is called a comparison model identifier, and the terminal can display multiple comparison model identifiers selected based on the first multiple selection operation. After determining multiple comparison model identifiers, the user can trigger a multi-model running operation, for example, the user can click the "Run" button control, at which time the terminal can respond to the multi-model running operation, and input the model input data corresponding to the model prompt and the target input data identifier into the large model corresponding to each comparison model identifier, so that each large model outputs the result and obtains the multi-model processing result. The terminal can display the multi-model processing result, so that the user can determine the effect of the model prompt based on the multi-model processing result.

[0126] In one embodiment, after the model prompt passes the evaluation, that is, after the model prompt does not meet the tuning conditions, a multi-model comparison operation may be performed on the model prompt to obtain a multi-model processing result.

[0127] In one of the embodiments, the multi-model processing result includes large model output results corresponding to each of the multiple comparison model identifiers; displaying the multi-model processing result includes: displaying each large model output result in the multi-model processing result on the same screen.

[0128] Specifically, the multi-model processing result includes the large model output results corresponding to the multiple comparison model identifiers, wherein the large model output result refers to the processing result of the large model output. The terminal can display the output results of each large model in the multi-model processing result on the same screen. For example, refer to Fig.11 When the multiple comparison model identifiers include the model identifier of the large model A, the model identifier of the large model B and the model identifier of the large model C, the processing result 1101 output by the large model A, the processing result 1102 output by the large model B and the processing result 1103 output by the large model C can be displayed on the same screen. Fig.11A schematic diagram showing a plurality of processing results displayed on the same screen in an embodiment is shown.

[0129] In this embodiment, by displaying the output results of each large model in the multi-model processing results on the same screen, it is convenient for users to compare the output results of each large model, and then determine whether the model prompt needs to be adjusted again based on the comparison results. For example, when the output results of each large model are consistent, it can be considered that the model prompt is applicable to each large model and can guide each large model to output the expected results; if the output results of each large model are inconsistent, or only a small part of the output results of the large model are consistent with the standard processing results, it can be considered that the model prompt is not applicable to each large model, and the model prompt can be adjusted again.

[0130] In one of the embodiments, the method further includes: in response to a batch processing operation for a model prompt, displaying a batch processing page, and displaying a multi-model selection area and a data batch input area in the batch processing page; in response to a trigger operation for the multi-model selection area, displaying multiple third candidate model identifiers; in response to a second multiple-selection operation for multiple third candidate model identifiers, displaying multiple batch processing model identifiers selected from the multiple third candidate model identifiers through the second multiple-selection operation; in response to a model input data batch input operation triggered in the data batch input area, displaying a target data set identifier input through the model input data batch input operation; in response to the batch processing operation, displaying batch processing results of batch processing of the model input data set corresponding to the target data set identifier by the large model corresponding to each of the multiple batch processing model identifiers based on the model prompt.

[0131] Specifically, when the user believes that the model prompt does not need to be tuned, the user can trigger a batch processing operation for the model prompt, and the terminal responds to the batch processing operation and displays a batch processing page. For example, the user can click the batch run button control in the operation control sequence, and then the terminal responds to the click operation on the batch run button control and displays a batch processing page. The batch processing page can display multiple model selection areas and data batch input areas. For example, refer to Fig.12 , Fig.12 1201 in the figure is the multi-model selection area. Fig.121202 is a data batch input area. When the user expects to select multiple large models through the multi-model selection area, the user can trigger the multi-model selection area. For example, the user can click the drop-down button control in the multi-model selection area, and then the terminal responds to the trigger operation on the multi-model selection area and displays multiple third candidate model identifiers. The user can select multiple model identifiers from multiple third candidate model identifiers, and the terminal responds to the second multiple selection operation on multiple third candidate model identifiers to determine the model identifier selected by the user from the multiple third candidate model identifiers, and the model identifier selected by the user from the multiple third candidate model identifiers is called a batch model identifier, and multiple batch model identifiers are displayed.

[0132] Furthermore, the user can batch input the model input data through the data batch input area. The terminal can display the data set identifier input through the model input data batch input operation in response to the model input data batch input operation triggered in the data batch input area. For example, a data batch input box can be displayed in the data batch input area. When the user clicks the data batch input box, the terminal can display multiple candidate data set identifiers, and the user can select one from them. Then, the terminal displays the data set identifier selected by the user as the target data set identifier in the data batch input box. The data set identifier is the identifier of the model input data set, and the model input data set can include multiple model input data.

[0133] Furthermore, the user can trigger a batch processing operation. For example, the user can click a "Run" button control, and the terminal responds to the batch processing operation and inputs the model input data set corresponding to the model prompt and the data set identifier into the large models corresponding to each batch model identifier, so that each large model outputs the results and obtains the batch processing results. Fig.12 A schematic diagram of a batch processing page in an embodiment is shown.

[0134] In one of the embodiments, the batch processing results include a result subset corresponding to each model input data in the model input data set corresponding to the target data set identifier; each result subset includes the result of the large model output corresponding to each batch model identifier; the batch processing results of batch processing the model input data set corresponding to the target data set identifier based on the model prompt by the large model corresponding to each of the multiple batch model identifiers are displayed, including: displaying each model input data in the model input data set corresponding to the target data set identifier; for each model input data in the model input data set corresponding to the target data set identifier, in the blank area corresponding to the model input data, displaying the result subset corresponding to the model input data.

[0135] The batch processing result includes the result subset corresponding to each model input data in the model input data set corresponding to the target data set identifier. For example, the model input data set corresponding to the target data set identifier includes model input data A, model input data B and model input data C, then the batch processing result includes result subset A corresponding to model input data A, result subset B corresponding to model input data B, and result subset C corresponding to model input data C. The result subset includes the results of the large model output corresponding to each batch model identifier, for example, result subset A includes the processing results of the large model corresponding to each batch model identifier based on the model prompt and the output of model input data A; result subset B includes the processing results of the large model corresponding to each batch model identifier based on the model prompt and the output of model input data B.

[0136] Specifically, when the batch processing result is obtained, the terminal can display each model input data in the model input data set corresponding to the target data set identifier, for example, Fig.12 , the terminal can display model input data A, model input data B, and model input data C. For each model input data in the model input data set, in the blank area corresponding to the model input data, the result subset corresponding to the model input data is displayed, for example, reference Fig.12 , display result subset A in the blank area corresponding to model input data A, display result subset B in the blank area corresponding to model input data B, and so on.

[0137] In the above embodiment, by providing a batch processing function, the large model can process multiple model input data in batches, thereby improving the model processing efficiency. By displaying the model input data and the result subsets in correspondence, it is convenient for users to correspond the model input data with the results output by the large model, thereby improving the user experience. By displaying the results output by multiple large models in the form of result subsets, it is convenient for users to determine the differences between the results output by each large model, improving the efficiency of determining the result differences, and improving the user experience.

[0138] In one of the embodiments, the method further includes: displaying a model binding control in response to a publishing operation on a model prompt; displaying multiple fourth candidate model identifiers in response to a triggering operation on a model binding control; binding a binding model identifier selected by a fourth selection operation from multiple fourth candidate model identifiers to the model prompt to obtain a binding result, and generating a publishing request based on the binding result, obtaining a binding result, and generating a publishing request based on the binding result; sending the publishing request to a terminal corresponding to a business approver; the sent publishing request is used to trigger the business approver to perform publishing approval based on the model prompt and the large model bound to the model prompt, and publish the model prompt if the approval is passed.

[0139] Specifically, when it is determined through batch model input data that the model prompt can guide the large model to output the expected result, the large model can be published so that other users can also use the model prompt. The user can trigger the model prompt publishing operation, for example, the user can click the publish button control, and then the terminal responds to the publishing operation for the model prompt and displays the model prompt publishing page. The model prompt publishing page can display a model binding control. When the user triggers the model binding control, for example, when the user clicks the model binding button control, the terminal can respond to the triggering operation for the model binding control and display multiple fourth candidate model identifiers, from which the user can select, and then the terminal responds to the user's fourth selection operation on the multiple fourth candidate model identifiers, determines the model identifier selected by the fourth selection operation, and calls the model identifier selected by the fourth selection operation a bound model identifier. The terminal binds the bound model identifier with the model prompt to obtain a binding result, and generates a publishing request based on the binding result.

[0140] Furthermore, the terminal may send the publishing request to the terminal corresponding to the business approver. When the terminal corresponding to the business approver receives the publishing request, the business approver may approve the publishing request. For example, the terminal corresponding to the business approver may determine the binding result carried in the publishing request, and determine the binding model identifier and model prompt in the binding result, and input the model prompt into the large model corresponding to the binding model identifier, and the terminal corresponding to the business approver may obtain the model input data, input the model input data into the large model, run the large model, and obtain the output result of the large model. The business approver may determine whether the output result of the large model is correct, and if it is correct, the approval is passed. If the approval is passed, the business approver may publish the model prompt online.

[0141] In one of the embodiments, an output result format confirmation area may also be displayed in the model prompt publishing page, and the user may confirm the format of the processing result output by the large model through the output result format confirmation area, and then the terminal may generate a publishing request based on the confirmed output result format and binding result.

[0142] In one embodiment, different fourth candidate model identifiers correspond to different version numbers.

[0143] In this embodiment, the model prompt is published online after approval, which can improve the security of the publication.

[0144] It is easy to understand that the above trigger operations include but are not limited to click operation, double-click operation, click and press operation, long press operation, etc.

[0145] In one embodiment, a self-service model prompt processing system is provided, through which business personnel can achieve one-stop self-service creation, adjustment and release of model prompts without relying on product development. Fig.13 , Fig.13 The overall process diagram of model prompt creation and release in one embodiment is shown. Business personnel can construct a model input data set, write model prompts, and debug the model prompts based on the model input data set. When debugging is successful, a release application can be created, and then the reviewer can confirm the effect of the model prompt when receiving the release application. After confirming the effect of the model prompt, the model prompt can be released online. Customers can use the published model prompts through the online business module. The online business model can call the large model, output business data based on the published model prompts used by the customer, and update the old business data based on the output business data.

[0146] In one embodiment, reference Fig.14 , Fig.14 The interactive diagram of self-service creation and publishing of model prompts in an embodiment is shown. After the business personnel pass the authority review and confirm their identity, they can self-service create the model prompts (also known as prompts) required by the business through the system in the corresponding business scenario. The business personnel provide model input data (also known as evaluation samples), input the created model prompts and model input data into the big model to obtain the processing results. The process of the big model outputting the processing results based on the model prompts and model input samples can also be regarded as evaluating the model prompts. Therefore, the business personnel can self-service evaluate the model prompts based on the model input data. The model prompt self-service processing system returns the processing results (also known as the evaluation results), and the business personnel analyze the processing results, optimize the model prompts as needed, and evaluate again. The big model returns the processing results (also known as the evaluation results) again. The business personnel cycle the above operations and perform multiple rounds of optimization. After the optimization results are passed, the business personnel preliminarily evaluate and confirm that the samples are available, expand the verification sample size, perform batch verification of the model input data set, and compare the effects of multiple models. The model prompt self-service processing system returns the batch verification results. The business personnel analyze the verification results, confirm that the results are available, confirm the selection of the large model version, output format and other details, and submit the release application. The business reviewer checks the model prompts and cross-validates that the data results are reasonable and available. After approval, the program automatically calls the model prompts, selects the confirmed large model, confirms the output format, and automatically releases it online for customer use.

[0147] In one embodiment, the model prompt self-service processing system is divided into: front-end application scenario layer, AI training / release middle platform layer, and back-end infrastructure layer. Fig.15, the front-end application scenario layer is directly oriented to user needs. Through scenario solutions such as information scenarios, content retrieval scenarios, customer service scenarios, and IT efficiency improvement scenarios, the scenarios are divided into three categories: AI for Business, AI for IT, and AI for Staff, to provide intelligent services for target users. AI for Business: The service is oriented to end users, providing business with content generation, service construction and other product supplies. For example, intelligent information writing can achieve low-cost and fast batch writing based on large models according to demand. AI for IT: It is aimed at internal R&D engineers to assist them in improving coding efficiency and quickly verifying and discovering defects, such as assisting in testing interfaces, conducting pipeline exception handling, and failure attribution. AI for Staff: It is aimed at internal business personnel to assist them in improving work efficiency. For example, abnormal article inspections can assist editors in identifying sensitive words, soft articles and other high-risk content in massive articles.

[0148] The AI ​​training / publishing layer of the middle platform includes the capability gateway, capability layer, thinking module and adapter. Among them, the capability gateway is the unified entrance of AI services, providing API calls and integration services. The capability layer includes model prompt writing / verification functions, open dialogue functions, data set / version management analysis functions and on-shelf release / system call functions. The model prompt writing / verification function is used to provide capabilities such as natural language model prompt writing and variable customization. The open dialogue function is used to achieve natural language interaction with users. The data set / version management analysis function is used to enter / modify the model input data set and analyze and compare the prompt effects of different versions of models. The on-shelf release / system call function is used for: model prompt release and formal call, historical version model prompt on-shelf release. The thinking module includes scenario design function, context memory function, domain knowledge function and capability plug-in. The scenario design function is used to customize AI application scenarios according to business needs. The context memory function is used to enable AI to understand and remember the conversation context. The domain knowledge function is used to integrate professional knowledge in a specific field. The capability plug-in is used to modularize AI capabilities for easy expansion and maintenance. The adapter includes large model adaptation function, data adaptation function and tool adaptation function. The big model adaptation function is used to ensure that the big model matches the business requirements. The data adaptation function is used to process and optimize data to fit the big model. The tool adaptation function is used for the entire third-party tools and systems. Fig.15 A schematic diagram of the architecture of a model-prompted self-service processing system in one embodiment is shown.

[0149] In one embodiment, the model prompt processing method is divided into three processes: model prompt writing, model prompt adjustment, and model prompt publishing. The overall process diagram of model prompt writing can be referred to Fig.16, step 1601 user accesses the system, step 1602, login authentication is required first, step 1603, if the permission has not been applied for, jump to the permission application process, the permissions are divided into business operators, business reviewers, and visitors, and the functions of different permissions are different. Business operators divide business lines according to business scenarios. Business operators in the same business scenario can view, create, modify, and run their own model prompts and model input data sets. Business operators across business scenarios can only browse and cannot perform other operations; business reviewers have the final release authority to effectively isolate and prevent risks; visitors can learn operations in the experience area and browse various business function areas, but cannot perform other operations; step 1604, if the permission has been opened, jump to the login authentication step, step 1605, if the user cancels the permission, the process ends. Step 1606, after the permission is opened, the system logs in again and further determines the permission type, step 1607, after the background system determines, it notifies the front-end routing to display the corresponding function, for example, for visitors, only the browse function is displayed. Step 1608: The business operator selects a business scenario. Step 1609: In the business scenario, a new model prompt is created. The model prompt is written in natural language. Step 1610: The business operator determines whether to import from an existing template. Step 1611: If it is imported from an existing template, a template can be selected. There are two sources of existing templates: Source 1: Templates created by the user; Source 2: Templates created and published by other users. Step 1612: After the business operator selects the template to import the model prompt, the content can be modified as appropriate. Step 1613: After the improvement, save it. The backend system stores the user input content. Step 1614: If the business operator chooses not to import from the template, enter the new creation process. In the front-end interface, fill in the model prompt name, model prompt description, model prompt body and other contents in turn, and determine the main quantitative content. Step 1615: Define the variable items in the model prompt body. Step 1616: Save it. The backend system stores the content entered by the business operator. At this time, the model prompt writing process is completed.

[0150] refer to Fig.17 , Fig.17The overall flow diagram of model prompt adjustment in one embodiment is shown. Step 1701, after the model prompt is created and written, the business operator can verify and adjust it. Select the model prompt to be adjusted, step 1702, enter the model input data in the page, step 1703, determine whether to import from the model input data set. There are two ways to input model input data: step 1703, method one: import the model input data set. The user uploads the sorted model input data table, and the system has agreed on the data cleaning and extraction method in advance to extract the data in the preset columns in the model input data table, for example, obtain the data in the "article title" column and obtain the data in the "article link" column, and use the data obtained from the model input data table as variables. Step 1705, method two: the business operator enters manually to obtain the variable part. Step 1706, after obtaining the variable, the variable can be replaced with the part marked with variable replacement in the model prompt to obtain the complete content source. Step 1707: The business operator selects a large model. Step 1708: The large model operation request is constructed, the large model interface is called, the complete content source is input into the large model, and the processing result returned by the large model is received. Step 1709: Based on the processing result returned by the large model, the business operator determines whether the model prompts that the single debugging has passed the evaluation. Step 1710: If it has not passed, it is optimized and retried. Step 1711: After the model prompts that the single evaluation has passed, a large number of samples are further verified for the model prompt. Step 1712: Determine whether the verification has passed. Step 1713: After the verification is passed, the application release process is followed and the application is released online.

[0151] refer to Fig.18 , Fig.18 A schematic diagram of the model prompt release in an embodiment is shown. After the large model training and debugging is completed, the business operator applies for the release process and submits the online approval form through the system interface. After the business auditor receives the approval form, he verifies that it is correct and agrees to approve it for release. After the large model is officially released, a version number is generated and submitted for release. If there is an online version online, the old version will be removed from the shelves and replaced with the new version. The business background triggers the call logic in the online production environment, obtains the RAG data content (Elasticsearch distributed search engine, vector database, distributed data warehouse), and constructs a model prompt call request by calling the online published and effective model prompt. The above request calls the large model interface agreed upon by the program, waits for the model to run, returns the generated results, and optimizes the output results and aggregates them for output.

[0152] The application of the above embodiments can produce the following beneficial effects: Promoting scientific and technological innovation and application: The model prompt self-service processing system can lower the application threshold of artificial intelligence technology, so that ordinary people who do not have an artificial intelligence technology background can also participate in the construction and optimization process of model prompts, thereby enhancing the vitality of scientific and technological innovation in artificial intelligence technology.

[0153] Improve business operation efficiency: By providing a one-click publishing function, the automated and intelligent process greatly shortens the construction, adjustment and publishing cycle of model prompts, and significantly improves the processing efficiency of model prompts.

[0154] Reduce business costs: Taking financial information services as an example, by building and using model prompts, the duplicate articles, soft articles, etc. in financial articles can be reduced based on large models and model prompts, thereby increasing the proportion of high-quality articles. Compared with manual screening of low-quality articles, it can also reduce labor costs, improve the efficiency of low-quality article screening, and reduce business costs.

[0155] Promote digital transformation: Promote the industry's business model from traditional to digital and intelligent, enhance industry competitiveness, and with the power of AI technology, business personnel can better understand customer needs, optimize service processes, and improve service quality, thereby enhancing market competitiveness. At the same time, the accurate prediction and decision support capabilities of large models will help business personnel gain a leading edge in risk management, product innovation, and other aspects.

[0156] In one specific embodiment, a model prompt processing method is provided, comprising:

[0157] The terminal displays a model prompting a login page of the self-service processing system, and in response to a login operation triggered in the login page, displays at least one business scenario and a model prompting a new control when the account role of the login account input in the login operation is the target role.

[0158] In response to a first selection operation for at least one business scenario, the terminal distinguishably displays the target business scenario selected by the first selection operation; in response to a trigger operation for creating a new model prompt control, when the logged-in account has the authority to create a model prompt under the target business scenario, the terminal displays the model prompt processing page under the target business scenario, and displays the model prompt input area, model input data configuration area and large model configuration area in the model prompt processing page.

[0159] In response to a model prompt input operation triggered in the model prompt input area, the terminal displays a model prompt input based on the model prompt input operation input.

[0160] In response to a model input data configuration operation triggered in the model input data configuration area, the terminal displays a target input data identifier configured based on the model input data configuration operation.

[0161] In response to the large model configuration operation triggered in the large model configuration area, the terminal displays the target model identifier configured based on the large model configuration operation; in response to the large model running operation, the model prompt and the model input data corresponding to the target input data identifier are input into the large model corresponding to the target model identifier.

[0162] The terminal calls the large model corresponding to the target model identifier, so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier based on the model prompt, obtains the processing result, and displays the processing result.

[0163] When it is determined based on the processing results that the model prompt meets the tuning conditions, the terminal responds to the tuning operation on the model prompt, redisplays the model prompt processing page, and redisplays the model prompt in the redisplayed model prompt processing page; responds to the modification operation on the redisplayed model prompt, displays the modified model prompt; uses the modified model prompt as the new model prompt, and returns to the step of responding to the large model running operation to continue execution.

[0164] In response to a trigger operation on the details control, the terminal displays a multi-model comparison page and displays a multi-model comparison control on the multi-model comparison page; in response to a trigger operation on the multi-model comparison control, multiple second candidate model identifiers are displayed; in response to a first multiple selection operation on multiple second candidate model identifiers, multiple comparison model identifiers selected from the multiple second candidate model identifiers based on the first multiple selection operation are displayed.

[0165] The terminal responds to the multi-model running operation and displays the multi-model processing results; the multi-model processing results are the results of processing the model input data corresponding to the target input data identifier based on the model prompt, with the large model corresponding to each of the multiple comparison model identifiers.

[0166] In response to a batch processing operation on a model prompt, the terminal displays a batch processing page, and displays a multi-model selection area and a data batch input area in the batch processing page; in response to a trigger operation on the multi-model selection area, the terminal displays multiple third candidate model identifiers.

[0167] In response to the second multiple selection operation on the multiple third candidate model identifiers, the terminal displays multiple batch model identifiers selected from the multiple third candidate model identifiers through the second multiple selection operation.

[0168] In response to the model input data batch input operation triggered in the data batch input area, the terminal displays the target data set identifier input through the model input data batch input operation; in response to the batch processing operation, the terminal displays the batch processing results of the large model corresponding to each of the multiple batch processing model identifiers performing batch processing on the model input data set corresponding to the target data set identifier based on the model prompt.

[0169] In response to a publish operation on a model prompt, the terminal displays a model binding control; in response to a trigger operation on the model binding control, the terminal displays multiple fourth candidate model identifiers; in response to a fourth selection operation on multiple fourth candidate model identifiers, the terminal binds the binding model identifier selected from the multiple fourth candidate model identifiers by the fourth selection operation to the model prompt, obtains a binding result, and generates a publish request based on the binding result.

[0170] The terminal sends the publishing request to the terminal corresponding to the business approver; the published request sent is used to trigger the business approver to perform publishing approval based on the model prompt and the large model bound to the model prompt, and publish the model prompt if the approval is passed.

[0171] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0172] The present application also provides an application scenario, and the application scenario applies the above-mentioned model prompt processing method. Specifically, the application of the model prompt processing method in the application scenario is as follows:

[0173] Financial practitioners can create model prompts through the model prompt self-service processing system, and tune the created model prompts to obtain the final model prompts, and publish the final model prompts online. Then, customers can perform corresponding tasks through the published model prompts, such as checking for duplicate financial articles, analyzing stocks, summarizing financial articles, etc.

[0174] The above application scenarios are merely illustrative descriptions. It is to be understood that the application of the schedule synchronization method provided in each embodiment of the present application is not limited to the above scenarios.

[0175] Based on the same inventive concept, the embodiment of the present application also provides a model prompt processing device for implementing the model prompt processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more model prompt processing device embodiments provided below can refer to the limitations of the model prompt processing method above, and will not be repeated here.

[0176] In one embodiment, Fig.19 As shown, a model prompt processing device 1900 is provided, including: a page display module 1902, an edit configuration module 1904 and a result display module 1906, wherein:

[0177] The page display module 1902 is used to display the model prompt processing page, and display the model prompt input area, the model input data configuration area and the large model configuration area in the model prompt processing page;

[0178] The editing configuration module 1904 is used to display the model prompt input based on the model prompt input operation in response to the model prompt input operation triggered in the model prompt input area; to display the target input data identifier configured based on the model input data configuration operation in response to the model input data configuration operation triggered in the model input data configuration area; to display the target model identifier configured based on the large model configuration operation in response to the large model configuration operation triggered in the large model configuration area;

[0179] The result display module 1906 is used to respond to the large model running operation and display the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt.

[0180] In one embodiment, the model prompt processing device 1900 also includes an adjustment module for redisplaying the model prompt processing page in response to a tuning operation on the model prompt, and redisplaying the model prompt in the redisplayed model prompt processing page when it is determined based on the processing result that the model prompt meets the tuning conditions; displaying the modified model prompt in response to a modification operation on the redisplayed model prompt; using the modified model prompt as a new model prompt, and returning to the step of responding to the large model running operation to continue execution.

[0181] In one of the embodiments, the page display module 1902 is also used to display the login page of the model prompt self-service processing system, and in response to a login operation triggered in the login page, when the account role of the login account input in the login operation is the target role, display at least one business scenario and a model prompt new creation control; in response to a first selection operation for at least one business scenario, distinguishably display the target business scenario selected by the first selection operation; in response to a trigger operation for a model prompt new creation control, display a model prompt processing page under the target business scenario.

[0182] In one embodiment, the page display module 1902 is also used to respond to the trigger operation of creating a new control for the model prompt, and when the login account has the authority to create the model prompt under the target business scenario, display the model prompt processing page under the target business scenario.

[0183] In one embodiment, the target input data is identified as a data link of the model input data; the model input data configuration area includes a first data input box; the editing configuration module 1904 is also used to respond to a link input operation triggered in the first data input box, and display the data link of the model input data input based on the link input operation in the first data input box.

[0184] In one embodiment, the model input data configuration area includes a second data input box and a model input data selection control; the editing configuration module 1904 is also used to display a data list in response to a trigger operation on the model input data selection control; in response to a second selection operation on multiple input data identifiers in the data list, a target input data identifier selected by the second selection operation from the multiple input data identifiers is displayed in the second data input box.

[0185] In one embodiment, the large model configuration area includes a model identification display box; the model identification display box includes a drop-down control; the editing configuration module 1904 is also used to display multiple first candidate model identifications in response to a trigger operation on the drop-down control; in response to a third selection operation on the multiple first candidate model identifications, display a target model identification selected from the multiple first candidate model identifications based on the third selection operation.

[0186] In one of the embodiments, the editing configuration module 1904 is also used to display a model prompt template list, which includes at least one template identifier and an operation control sequence corresponding to each template identifier; in response to a triggering operation on an editing control in the operation control sequence, determining a triggered template identifier corresponding to the triggered editing control; displaying a triggered model prompt template corresponding to the triggered template identifier; and obtaining a model prompt in response to a modification operation on the triggered model prompt template.

[0187] In one of the embodiments, the result display module 1906 is also used to respond to the large model running operation, input the model input data corresponding to the model prompt and the target input data identifier into the large model corresponding to the target model identifier; call the large model corresponding to the target model identifier so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier based on the model prompt, obtains the processing result, and displays the processing result.

[0188] In one of the embodiments, the result display module 1906 is also used to display a processing result display page, and display the processing results and model prompts in the processing result display page; based on the correlation between the processing results and each sentence in the model prompt, the sentences in the model prompts in the processing result display page are differentiated and displayed; model prompt optimization information is generated based on the differentiated and displayed sentences, and the model prompt optimization information is displayed.

[0189] In one of the embodiments, a save control is displayed in the model prompt processing page; the model prompt processing device 1900 also includes a save module, which is used to generate a target model prompt template and a target template identifier corresponding to the target model prompt template based on the model prompt in response to a trigger operation on the save control; in the model prompt template list under the business scenario corresponding to the model prompt, the target template identifier and the operation control sequence corresponding to the target template identifier are displayed.

[0190] In one of the embodiments, the operation control sequence corresponding to the target template identifier also includes a detail control; the model prompt processing device 1900 also includes a multi-model comparison module, which is used to display a multi-model comparison page in response to a trigger operation on the detail control, and display a multi-model comparison control in the multi-model comparison page; in response to a trigger operation on the multi-model comparison control, multiple second candidate model identifiers are displayed; in response to a first multiple-selection operation on multiple second candidate model identifiers, multiple comparison model identifiers selected from multiple second candidate model identifiers based on the first multiple-selection operation are displayed; in response to a multi-model running operation, a multi-model processing result is displayed; the multi-model processing result is the result of processing the model input data corresponding to the target input data identifier by the large model corresponding to each of the multiple comparison model identifiers based on the model prompt.

[0191] In one of the embodiments, the multi-model processing result includes large model output results corresponding to each of the multiple comparison model identifiers; the multi-model comparison module is also used to display the output results of each large model in the multi-model processing result on the same screen.

[0192] In one embodiment, the model prompt processing device 1900 also includes a batch running module, which is used to display a batch processing page in response to a batch processing operation on the model prompt, and display a multi-model selection area and a data batch input area in the batch processing page; in response to a trigger operation on the multi-model selection area, display multiple third candidate model identifiers; in response to a second multiple selection operation on the multiple third candidate model identifiers, display multiple batch processing model identifiers selected from the multiple third candidate model identifiers through the second multiple selection operation; in response to a model input data batch input operation triggered in the data batch input area, display the target data set identifier input through the model input data batch input operation; in response to the batch processing operation, display the batch processing results of the large model corresponding to each of the multiple batch processing model identifiers performing batch processing on the model input data set corresponding to the target data set identifier based on the model prompt.

[0193] In one of the embodiments, the batch processing results include result subsets corresponding to each model input data in the model input data set corresponding to the target data set identifier; each result subset includes the results of the large model output corresponding to each batch processing model identifier; the batch running module is also used to display each model input data in the model input data set corresponding to the target data set identifier; for each model input data in the model input data set corresponding to the target data set identifier, the result subset corresponding to the model input data is displayed in the blank area corresponding to the model input data.

[0194] In one embodiment, the model prompt processing device 1900 also includes a publishing application module, which is used to display a model binding control in response to a publishing operation on the model prompt; display multiple fourth candidate model identifiers in response to a trigger operation on the model binding control; in response to a fourth selection operation on multiple fourth candidate model identifiers, bind the binding model identifier selected by the fourth selection operation from the multiple fourth candidate model identifiers with the model prompt to obtain a binding result, and generate a publishing request based on the binding result; send the publishing request to the terminal corresponding to the business approver; the sent publishing request is used to trigger the business approver to perform publishing approval based on the model prompt and the large model bound to the model prompt, and publish the model prompt if the approval is passed.

[0195] Each module in the above-mentioned model prompt processing device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0196] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 20 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a model prompt processing method is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0197] Those skilled in the art will understand that Fig. 20 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0198] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0199] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0200] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the above-mentioned method embodiments.

[0201] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0202] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0203] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0204] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A model prompt processing method, characterized in that: The method comprises: Displaying a model prompt processing page, and displaying a model prompt input area, a model input data configuration area, and a large model configuration area in the model prompt processing page; In response to a model prompt input operation triggered in the model prompt input area, displaying a model prompt input based on the model prompt input operation; In response to a model input data configuration operation triggered in the model input data configuration area, displaying a target input data identifier configured based on the model input data configuration operation; In response to a large model configuration operation triggered in the large model configuration area, displaying a target model identifier configured based on the large model configuration operation; In response to the large model running operation, the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt is displayed.

2. The method according to claim 1, characterized in that The method further comprises: In a case where it is determined based on the processing result that the model prompt meets the tuning condition, in response to a tuning operation on the model prompt, the model prompt processing page is redisplayed, and the model prompt is redisplayed in the redisplayed model prompt processing page; In response to a modification operation on the re-displayed model prompt, displaying a modified model prompt; The modified model prompt is used as a new model prompt, and the step of responding to the large model running operation is returned to continue execution.

3. The method according to claim 1, characterized in that The display model prompts the processing page, including: Displaying a login page of a model prompt self-service processing system, and in response to a login operation triggered in the login page, displaying at least one business scenario and a model prompt new creation control when the account role of the login account input in the login operation is the target role; In response to a first selection operation for the at least one business scenario, distinctively displaying a target business scenario selected by the first selection operation; In response to a triggering operation of creating a new control for the model prompt, a model prompt processing page for the target business scenario is displayed.

4. The method according to claim 3, characterized in that In response to the triggering operation of creating a new control for the model prompt, displaying the model prompt processing page under the target business scenario includes: In response to a trigger operation for creating a new control for the model prompt, when the login account has the authority to create a model prompt under the target business scenario, a model prompt processing page under the target business scenario is displayed.

5. The method according to claim 1, characterized in that The target input data identifier is a data link of the model input data; the model input data configuration area includes a first data input box; The displaying of the target input data identifier configured based on the model input data configuration operation in response to the model input data configuration operation triggered in the model input data configuration area comprises: In response to a link input operation triggered in the first data input box, a data link of the model input data input based on the link input operation is displayed in the first data input box.

6. The method according to claim 1, characterized in that The model input data configuration area includes a second data input box and a model input data selection control; The displaying of the target input data identifier configured based on the model input data configuration operation in response to the model input data configuration operation triggered in the model input data configuration area comprises: In response to a triggering operation of inputting a data selection control for the model, displaying a data list; In response to a second selection operation on a plurality of input data identifiers in the data list, a target input data identifier selected by the second selection operation from the plurality of input data identifiers is displayed in the second data input box.

7. The method according to claim 1, characterized in that The large model configuration area includes a model identification display box; the model identification display box includes a drop-down control; In response to the large model configuration operation triggered in the large model configuration area, displaying the target model identifier configured based on the large model configuration operation includes: In response to a trigger operation on the drop-down control, displaying a plurality of first candidate model identifiers; In response to a third selection operation on the plurality of first candidate model identifiers, a target model identifier selected from the plurality of first candidate model identifiers based on the third selection operation is displayed in the model identifier display frame.

8. The method according to claim 1, characterized in that The method further comprises: Displaying a model prompt template list, wherein the model prompt template list includes at least one template identifier and an operation control sequence corresponding to each of the template identifiers; In response to a triggering operation on an edit control in the operation control sequence, determining a triggered template identifier corresponding to the triggered edit control; Displaying the triggered model prompt template corresponding to the triggered template identifier; In response to a modification operation on the triggered model prompt template, a model prompt is obtained.

9. The method according to claim 1, characterized in that: In response to the large model running operation, displaying the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt, includes: In response to a large model running operation, inputting the model prompt and the model input data corresponding to the target input data identifier into the large model corresponding to the target model identifier; The large model corresponding to the target model identifier is called so that the large model corresponding to the target model identifier processes the model input data corresponding to the target input data identifier based on the model prompt, obtains a processing result, and displays the processing result.

10. The method according to claim 9, characterized in that The displaying of the processing result includes: Displaying a processing result display page, and displaying the processing result and the model prompt on the processing result display page; According to the correlation between the processing result and each sentence in the model prompt, the sentences in the model prompt in the processing result display page are distinguished and displayed; Model prompt optimization information is generated based on the differentiated and displayed sentences, and the model prompt optimization information is displayed.

11. The method according to claim 1, characterized in that: The model prompt processing page displays a save control; the method further includes: In response to a trigger operation on the save control, generating a target model prompt template and a target template identifier corresponding to the target model prompt template based on the model prompt; In the model prompt template list under the business scenario corresponding to the model prompt, the target template identifier and the operation control sequence corresponding to the target template identifier are displayed.

12. The method according to claim 11, characterized in that The operation control sequence corresponding to the target template identifier also includes a detail control; the method further includes: In response to a triggering operation on the detail control, displaying a multi-model comparison page, and displaying a multi-model comparison control in the multi-model comparison page; In response to a triggering operation on the multi-model comparison control, displaying a plurality of second candidate model identifiers; In response to a first multiple selection operation on the plurality of second candidate model identifiers, displaying a plurality of comparison model identifiers selected from the plurality of second candidate model identifiers based on the first multiple selection operation; In response to the multi-model running operation, the multi-model processing result is displayed; the multi-model processing result is the result of processing the model input data corresponding to the target input data identifier based on the model prompt by the large model corresponding to each of the multiple comparison model identifiers.

13. The method according to claim 12, characterized in that The multi-model processing result includes the large model output results corresponding to each of the multiple comparison model identifiers; The display of the multi-model processing results includes: The output results of each large model in the multi-model processing results are displayed on the same screen.

14. The method according to claim 1, characterized in that The method further comprises: In response to the batch processing operation prompted for the model, displaying a batch processing page, and displaying a multi-model selection area and a batch data input area in the batch processing page; In response to a trigger operation on the multi-model selection area, displaying a plurality of third candidate model identifiers; In response to a second multiple selection operation on the plurality of third candidate model identifiers, displaying a plurality of batch model identifiers selected from the plurality of third candidate model identifiers through the second multiple selection operation; In response to a model input data batch input operation triggered in the data batch input area, displaying a target data set identifier input by the model input data batch input operation; In response to a batch processing operation, batch processing results of the large model corresponding to each of the multiple batch processing model identifiers performing batch processing on the model input data set corresponding to the target data set identifier based on the model prompt are displayed.

15. The method according to claim 14, characterized in that The batch processing result includes a result subset corresponding to each model input data in the model input data set corresponding to the target data set identifier; each of the result subsets includes a result of a large model output corresponding to each batch processing model identifier; The display of the batch processing results of the large model corresponding to each of the multiple batch processing model identifiers batch processing the model input data set corresponding to the target data set identifier based on the model prompt includes: Display each model input data in the model input data set corresponding to the target data set identifier; Each model input data in the model input data set corresponding to the target data set is identified, and a result subset corresponding to the model input data is displayed in a blank area corresponding to the model input data.

16. The method according to claim 1, characterized in that The method further comprises: In response to a publish operation for the model prompt, displaying a model binding control; In response to a trigger operation on the model binding control, displaying a plurality of fourth candidate model identifiers; In response to a fourth selection operation on the plurality of fourth candidate model identifiers, binding a binding model identifier selected from the plurality of fourth candidate model identifiers by the fourth selection operation with the model prompt to obtain a binding result, and generating a publishing request based on the binding result; The publishing request is sent to the terminal corresponding to the business approver; the sent publishing request is used to trigger the business approver to perform publishing approval based on the model prompt and the large model bound to the model prompt, and publish the model prompt if the approval is passed.

17. A model prompt processing device, characterized in that: The device comprises: A page display module, used to display a model prompt processing page, and display a model prompt input area, a model input data configuration area and a large model configuration area in the model prompt processing page; an editing configuration module, for, in response to a model prompt input operation triggered in the model prompt input area, displaying a model prompt input based on the model prompt input operation; in response to a model input data configuration operation triggered in the model input data configuration area, displaying a target input data identifier configured based on the model input data configuration operation; in response to a large model configuration operation triggered in the large model configuration area, displaying a target model identifier configured based on the large model configuration operation; A result display module is used to respond to the large model running operation and display the processing result of the large model corresponding to the target model identifier processing the model input data corresponding to the target input data identifier according to the guidance of the model prompt.

18. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 16 are implemented.

19. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 16 are implemented.

20. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 16 are implemented.

Images