Information processing method and related device

Abstract

The invention discloses an information processing method and a related device, and provides a multi-layer initial instruction of a domain specific language, the i layer instruction in the multi-layer initial instruction comprises an i corresponding instruction, the multi-layer initial instruction is refined layer by layer from top to bottom, the i model instance is determined according to the i corresponding instruction, and the i model instance is determined according to the i corresponding instruction. The ith corresponding instruction is an instruction of the ith layer in the multiple layers of initial instructions and is a subordinate instruction of the (i-1) th prediction instruction, and the initial instructions of all the layers are mutually independent. Response information is determined according to the demand information through the multiple large model instances, an ith prediction instruction is generated through the ith large model instance in the multiple large model instances, if the execution result of the ith prediction instruction is matched with the demand information, the execution result of the ith prediction instruction serves as the response information, and a first prediction instruction is generated according to the demand information; when i is greater than or equal to 2, the ith prediction instruction is generated according to the (i-1) th prediction instruction, so that the instruction determination accuracy of each large model instance is improved, and the response information determination accuracy is further improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an information processing method and related devices. Background Art

[0002] Low-code development is a way of developing applications. It can quickly develop applications through visualization technology and writing a small amount of logical code. The platform used to implement low-code development is called a low-code platform. The developed application can include instructions to be executed. The low-code platform can execute the instructions to be executed to run the application and obtain execution results.

[0003] Currently, the coding ability of the big model can be used to directly write code to obtain instructions to be executed, and the low-code platform runs the instructions to be executed to achieve application development. However, the corpus of the big model is a general code, and the low-code platform is usually a platform for a specific field. The big model cannot understand the capabilities of the platform for a specific field, and the instructions it outputs can only achieve more common functions in the industry, such as certain tool functions or general components, etc., but cannot combine or call the advantages and capabilities of the platform for a specific field, and the functional scope is relatively narrow. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides an information processing method and related devices, which integrate the semantics of multi-layer initial instructions into multiple large model instances respectively, can combine the advantages and capabilities of specific domain languages, reduce the global dependence of each large model instance, improve the instruction determination accuracy of each large model instance, and improve the demand response efficiency.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In one aspect, the present application provides an information processing method, the method comprising:

[0007] Obtain demand information;

[0008] Determine response information according to the demand information through multiple large model instances; in the process of determining the response information, generate an i-th prediction instruction through the i-th large model instance among the multiple large model instances, and if the execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer;

[0009] Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

[0010] On the other hand, the present application provides an information processing device, the device comprising:

[0011] An information acquisition unit, used for acquiring demand information;

[0012] an information processing unit, configured to determine response information according to the demand information through a plurality of large model instances; in the process of determining the response information, generate an i-th prediction instruction through an i-th large model instance among the plurality of large model instances, and if an execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer;

[0013] Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

[0014] In another aspect, the present application provides a computer device, the device comprising a processor and a memory:

[0015] The memory is used to store a computer program and transmit the computer program to the processor;

[0016] The processor is used to execute the information processing method described in the above aspect according to the instructions in the computer program.

[0017] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the information processing method described in the above aspects.

[0018] On the other hand, an embodiment of the present application provides a computer program product including a computer program, which, when executed on a computer device, enables the computer device to execute the information processing method.

[0019] It can be seen from the above technical solution that a multi-layer initial instruction of a domain-specific language is provided, and the i-th layer instruction in the multi-layer initial instruction includes the i-th corresponding instruction. The initial instruction at the i-th layer has detailed information relative to the initial instruction at the i-1th layer, and i is a positive integer, that is, the multi-layer initial instruction is refined layer by layer from top to bottom, so that the operation content of the multi-layer initial instruction is clarified layer by layer. The i-th large model instance is determined according to the i-th corresponding instruction. The i-th corresponding instruction is the instruction of the i-th layer in the multi-layer initial instruction, and is a subordinate instruction of the i-1th prediction instruction. In this way, the semantics of the multi-layer initial instruction is integrated into the large model instance, so that the response process to the demand information can be combined with the advantages and capabilities of the specific domain language, which is conducive to the expansion of functions. Different large model instances can use different initial instructions, and the initial instructions of different layers are used to determine different large model instances. The initial instructions of each layer are independent of each other and there is no problem of mutual interference, which is conducive to improving the response accuracy of each large model instance.

[0020] After obtaining the demand information, the response information can be determined according to the demand information through multiple large model instances. In the process of determining the response information, the i-th prediction instruction can be generated through the i-th large model instance in the multiple large model instances. If the execution result of the i-th prediction instruction matches the demand information, the execution result of the i-th prediction instruction is used as the response information, that is, the response to the demand information can be realized by using multiple large model instances. The multiple large model instances work together to improve the accuracy and efficiency of the demand response. Among them, the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction. The generated i-th prediction instruction is related to the i-th corresponding instruction, so that the i-th prediction instruction is the subordinate instruction of the i-1-th prediction instruction, that is, after the first large model instance determines the prediction instruction according to the demand information, other large model instances are used to further refine the prediction instruction, so that the gradually refined prediction instructions are obtained through multiple large model instances. A single large model instance only needs to ensure its own instruction decision, which reduces the global dependence of each large model instance, improves the instruction determination accuracy of each large model instance, and then improves the determination accuracy of the response information, and improves the overall stability and accuracy of the information processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A schematic diagram of an application scenario of an information processing method provided in an embodiment of the present application;

[0023] Figure 2 A flowchart of an information processing method provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of a graphical user interface of a low-code platform provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of a graphical user interface of another low-code platform provided in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of a hierarchical structure of a DSL provided in an embodiment of the present application;

[0027] Figure 6 A schematic diagram of the working process of multiple large model instances provided in an embodiment of the present application;

[0028] Figure 7 A schematic diagram of the working process of another multiple large model instance provided in an embodiment of the present application;

[0029] Figure 8 A schematic diagram of a graphical user interface of another low-code platform provided in an embodiment of the present application;

[0030] Fig. 9 A structural block diagram of an information processing device provided in an embodiment of the present application;

[0031] Fig.10 A structural diagram of a terminal device provided in an embodiment of the present application;

[0032] Fig.11 A structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] The embodiments of the present application are described below in conjunction with the accompanying drawings.

[0034] Currently, the coding ability of large models is used to directly write codes to obtain instructions to be executed, and the low-code platform runs the instructions to be executed to achieve application development. However, the corpus of large models is a universal code, and low-code platforms are usually domain-specific platforms. Large models cannot understand the capabilities of domain-specific platforms, and the instructions they output can only achieve more common functions in the industry, but cannot combine or call on the advantages and capabilities of domain-specific platforms, and the functional scope is relatively narrow.

[0035] In order to solve the above technical problems, the present application provides an information processing method and related devices, which integrate the semantics of multi-layer initial instructions into multiple large model instances respectively, can combine the advantages and capabilities of specific domain languages, reduce the global dependence of each large model instance, improve the instruction determination accuracy of each large model instance, and improve the demand response efficiency.

[0036] The information processing method provided in the embodiments of the present application can be implemented by a computer device, which can be a terminal device or a server, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal devices include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc. Terminal devices and servers can be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0037] In order to facilitate understanding of the technical solution provided by the present application, an information processing method provided by an embodiment of the present application will be introduced below in combination with a practical application scenario.

[0038] Figure 1 A schematic diagram of an application scenario of an information processing method provided by an embodiment of the present application is shown, in which a server 100 and a terminal device 200 are included, and an application for information processing is installed in the terminal device 200, and the server 100 and the terminal device 200 corresponding to the application interact through a network. The server 100 or the terminal device 200 can be used as the aforementioned computer device to perform information processing according to the demand information, and the terminal device 200 is used to interact with the user, which can be used to obtain the demand information and display the information processing results. The following will be described by taking the server 100 as the aforementioned computer device as an example.

[0039] The server 100 provides multi-layer initial instructions in a domain-specific language. The i-th layer of instructions in the multi-layer initial instructions includes the i-th corresponding instructions. The initial instructions at the i-th layer have detailed information relative to the initial instructions at the i-1th layer. i is a positive integer, that is, the multi-layer initial instructions are refined layer by layer from top to bottom, so that the operation content of the multi-layer initial instructions is clarified layer by layer. The i-th large model instance is determined according to the i-th corresponding instruction. The i-th corresponding instruction is the instruction of the i-th layer in the multi-layer initial instructions, and is the lower-level instruction of the i-1th prediction instruction. In this way, the semantics of the multi-layer initial instructions are integrated into the large model instance, so that the response process to the demand information can be combined with the advantages and capabilities of the specific domain language, which is conducive to the expansion of functions. Different large model instances can use different initial instructions. Initial instructions at different layers are used to determine different large model instances. The initial instructions at each layer are independent of each other and there is no problem of mutual interference, which is conducive to improving the response accuracy of each large model instance.

[0040] After obtaining the demand information, the server 100 can determine the response information according to the demand information through multiple large model instances. In the process of determining the response information, the ith prediction instruction can be generated through the ith large model instance among the multiple large model instances. If the execution result of the ith prediction instruction matches the demand information, the execution result of the ith prediction instruction is used as the response information, where i is a positive integer. That is, multiple large model instances can be used to respond to the demand information. The multiple large model instances work together to improve the accuracy and efficiency of the demand response. Among them, the first prediction instruction is generated according to the demand information, and the i-th prediction instruction is generated according to the i-1-th prediction instruction when i≥2. The generated i-th prediction instruction is related to the i-th corresponding instruction, so that the i-th prediction instruction is a subordinate instruction of the i-1-th prediction instruction. That is, after the first large model instance determines the prediction instruction according to the demand information, other large model instances are used to further refine the prediction instruction, so as to obtain gradually refined prediction instructions through multiple large model instances. A single large model instance only needs to ensure its own instruction decision, which reduces the global dependence of each large model instance, improves the instruction determination accuracy of each large model instance, and then improves the determination accuracy of the response information, and improves the overall stability and accuracy of the information processing process.

[0041] Figure 2 This is a flowchart of an information processing method provided in an embodiment of the present application. In this embodiment, a server is used as the aforementioned computer device for illustration. The information processing method may include:

[0042] S101, obtaining demand information.

[0043] In an embodiment of the present application, demand information can be obtained. The demand information is information that reflects the user's needs. It can be information input by the user or information obtained based on the information input by the user. In the application editing stage of the low-code platform, users develop applications through the low-code platform. Specifically, the low-code platform can display a graphical user interface (GUI) to the user, so that the demand information can be obtained through the graphical user interface to develop applications that match the demand information. The graphical user interface can also display demand matching instructions, execution process prompt information, etc. The low-code platform can also execute the application that matches the demand information after application development to obtain response information corresponding to the demand information, and display the response information corresponding to the demand information.

[0044] Specifically, the demand information can be obtained through the terminal device, and the terminal information can obtain the demand information in response to the trigger operation of the information acquisition control, wherein the information acquisition control can be a send control or a submit control. The demand information can be obtained through the input box control. In specific implementation, the demand information can be obtained through the input box control in response to the trigger operation of the information acquisition control. The information acquisition control and the input box control can be displayed in a graphical user interface, and the graphical user interface can be displayed in the form of a chat box; after obtaining the demand information, the demand information or user input content can also be displayed, and the demand information or user input content can be displayed in the chat box as the content of the chat bubble to improve the interactive fun.

[0045] refer to Figure 3 As shown, a schematic diagram of a graphical user interface of a low-code platform provided in an embodiment of the present application, the low-code platform is identified in the form of an artificial intelligence (AI) assistant, the graphical user interface includes a chat box 10, an input box control 110 and a send control 111, the chat box 10 displays a platform logo 120 and a platform prompt bubble 121, the content of the platform logo 120 includes AI, the content of the platform prompt bubble 121 is: "Welcome to the AI ​​assistant, I can automatically realize your development needs, how can I help you?", the input box control 110 displays a prompt content: "Please enter your requirements, such as "generate a bar chart.", when there is input content in the input box control 110, the prompt content is hidden, and when the send control 111 is triggered, the demand information can be determined according to the input content of the input box control 110.

[0046] refer to Figure 4As shown, a schematic diagram of a graphical user interface of another low-code platform provided in an embodiment of the present application is shown in the chat box 10. A user identifier 130 and a user chat bubble 131 are displayed. The content of the user chat bubble 131 is "Please help me generate a bar chart to count the number of male and female merchants." This content is used as input content in the input box control 110. Based on this input content, the demand information can be obtained as: "Generate a bar chart to count the number of male and female merchants."

[0047] S102, determining response information according to the demand information through multiple large model instances.

[0048] In an embodiment of the present application, multiple layers of initial instructions in a domain-specific language can be provided, and the large model can be adjusted using the multiple layers of initial instructions to obtain multiple large model instances.

[0049] Domain-specific language (DSL), also known as domain-specific language, refers to a computer language declared to solve specific problems in a specific field. Low-code platforms can implement corresponding functions by executing domain-specific languages. DSL can have a general language syntax or a custom syntax. DSLs include Hyper Text Markup Language (HTML) for displaying web pages and Structured Query Language (SQL) for database query and programming.

[0050] A large model refers to a deep learning model with huge parameter scale, massive training data, and high computing resource requirements, such as a large language model (LLM). The large language model is used to process natural language related tasks. It can be used for low-code development, thereby converting natural language demand information into executable instructions.

[0051] If the DSL is flattened and input into the big model to train the big model instance, the big model instance can get the DSL as an output instruction when used in low-code development. However, the process of training the big model instance requires a huge amount of calculation. If the pre-trained big model is fine-tuned using the DSL to get the big model instance, the DSL can also be obtained as an output instruction. However, there are problems of high cost and high bit error rate. This is because a complete set of DSL specifications and the corpus described by it are very large. Even fine-tuning the big model requires a very high cost. In addition, since only the big model is fine-tuned, the big model may not be able to fully understand the DSL. Different DSLs are easy to "interfere" with each other, causing the big model to have "hallucinations", resulting in the big model instance being unable to output a DSL that meets the platform specifications every time. DSLs that do not meet the platform specifications cannot be run by the low-code platform, affecting the stability and availability of the information processing system.

[0052] Therefore, in the embodiment of the present application, the DSL is encapsulated into a multi-layer initial instruction, that is, a multi-layer initial instruction of the DSL can be provided, a layer of initial instructions may include one or more initial instructions, an initial instruction may be an atomic instruction of the platform, the initial instruction located at the i-th layer is the lower layer instruction of the initial instruction located at the i-1th layer, i is a positive integer, and the total number of layers of multi-layer initial instructions can be recorded as n, then n≥i. In other words, the DSL in the embodiment of the present application can be designed as a multi-layer structure based on chain thinking, encapsulated as a platform atomic initial instruction, and obtain a multi-layer initial instruction. The use of multi-layer initial instructions to realize the response to the demand information enables the response process to the demand information to be combined with the advantages and capabilities of the language in a specific field, which is conducive to the expansion of functions.

[0053] The multiple layers of initial instructions form a tree structure, the initial instructions at the i-th layer are the lower-level instructions of the initial instructions at the i-1th layer, that is, the lower-level initial instructions are the lower-level instructions of the upper-level initial instructions, the initial instructions at the i-th layer have detailed information relative to the initial instructions at the i-1th layer, that is, the lower-level instructions have detailed information relative to the upper-level instructions, that is, the DSL in the embodiment of the present application is hierarchical, and the multiple layers of initial instructions are refined layer by layer from top to bottom, so that the operation content of the multiple layers of initial instructions is clarified layer by layer. An initial instruction can have one lower-level instruction or multiple lower-level instructions, and the multiple lower-level instructions respectively have their own lower-level instructions, so that the upper-level initial instructions to the lower-level initial instructions extend and diffuse in a chain, and the respective lower-level instructions of the multiple initial instructions belonging to the same layer can be located at the same layer of the multiple layers of initial instructions, that is, the multiple initial instructions at the same layer can be divided into multiple groups, each group is a lower-level instruction of multiple upper-level initial instructions.

[0054] Among them, an initial instruction can be a generation instruction or an acquisition instruction, and the multi-layer initial instruction includes at least one of the generation instruction and the acquisition instruction. If the target instruction is a generation instruction, the detailed information can be at least one of the generation object information and the generation method information, the generation object information such as the type of the generated object, the generation method information such as the storage location or related network interface of the generated object, etc. If the target instruction is an acquisition instruction, the detailed information can be at least one of the acquisition object information and the acquisition method information, the acquisition object information such as the type of the acquired object, etc., the acquisition method information such as the storage location or related network interface of the acquired object, etc.

[0055] refer to Figure 5 As shown, it is a schematic diagram of the hierarchical structure of a DSL provided by an embodiment of the present application. The DSL includes multiple layers of initial instructions. From the figure, it can be seen that the higher the initial instructions, the more abstract they are, and the larger the functional periphery, while the initial instructions of the lower layers will gradually refine the operations. Among the multiple layers of initial instructions, the first layer of initial instructions includes generation class instructions and acquisition class instructions. The subordinate instructions of the generation class instructions are located in the second layer, including generating data, generating UI (User Interface) components and other generation instructions (indicated by "..."), etc. Among them, the lower-level instructions for generating data are located at the third layer, including generating dynamic data and generating static data, etc. The lower-level instructions for generating dynamic data are located at the fourth layer, including generating data based on the database and generating data based on the HTTP (Hypertext Transfer Protocol) interface, etc. The lower-level instructions for generating static data are also located at the fourth layer, including generating strings and generating integers, etc. The lower-level instructions for generating UI components are located at the third layer, including generating chart-like components and generating input box-like components, etc. The lower-level instructions for generating chart-like components are located at the fourth layer, including generating bar charts and generating pie charts, etc. The lower-level instructions for generating input box-like components are located at the fourth layer, including generating single-line input boxes, generating multi-line input boxes, generating selection input boxes, etc. The lower-level instructions of the acquisition-like instructions, such as obtaining UI information, obtaining data information and other acquisition instructions, the lower-level instructions for obtaining UI information obtain the information of a certain UI component, etc., which are not illustrated here.

[0056] In the embodiment of the present application, the i-th large model instance can be determined according to the i-th corresponding instruction, and the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the DSL, so that the 1st corresponding instruction is the initial instruction of the 1st layer in the multi-layer initial instructions, and when i≥2, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions, and is the subordinate instruction of the i-1th prediction instruction, so that the semantics of the multi-layer initial instructions are integrated into the large model instance, so that the response process to the demand information can be combined with the advantages and capabilities of the language in a specific field, which is conducive to the expansion of functions. When determining multiple large model instances in the large model structure, the i-th large model instance is trained according to the determined i-1th prediction instruction, so that the model training is targeted, and only the corresponding initial instructions need to be input for a specific large model instance, so as to realize the layer-by-layer input of the initial instructions, different large model instances can use different initial instructions, and the initial instructions of different layers are used to determine different large model instances. The initial instructions of each layer are independent of each other and there is no problem of mutual interference, which greatly reduces the possibility of "hallucination" of the large model caused by inputting too many initial instructions at one time, ensures the effective input of the initial instructions, and is conducive to improving the accuracy of instruction determination of each large model instance, thereby improving the overall stability and accuracy of the information processing process. This allows the big model to fully call on the capabilities of the low-code platform itself, while avoiding the big model outputting predictive instructions that the platform cannot understand. This greatly improves the accuracy and stability of the low-code platform in the process of automating user needs, greatly improving the efficiency of application development while lowering the development threshold.

[0057] Specifically, the model to be optimized can be adjusted according to the i-th corresponding instruction to obtain the i-th large model instance. The model to be optimized is obtained by pre-training the initial model, so that the i-th corresponding instruction is used to fine-tune the model to be optimized, reducing the amount of calculation in the instruction generation process. The model to be optimized is pre-trained and used as the base model of the large model. Multiple large model instances can be obtained by fine-tuning the same model to be optimized respectively, and have different functions due to their different model parameters and / or different model structures.

[0058] After obtaining the demand information, response information can be determined according to the demand information through multiple large model instances, and the i-th prediction instruction can be generated through the i-th large model instance among the multiple large model instances. If the execution result of the i-th prediction instruction matches the demand information, the execution result of the i-th prediction instruction is used as the response information, and i is a positive integer. That is, the entire automation development process is not completed by one large model instance, but by multiple large model instances working together. Multiple large model instances work together to complete a work task to respond to the demand information. The output of one large model instance can be used as the input of another large model instance. Of course, multiple large model instances can perform data processing at the same time to improve the accuracy and efficiency of determining the response information.

[0059] Among them, the first prediction instruction is generated according to the demand information, and the i-th prediction instruction is generated according to the i-1-th prediction instruction when i≥2. Since the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction. Therefore, in the process of the i-th large model instance determining the i-th prediction instruction according to the i-1-th prediction instruction, the i-1-th prediction instruction can be refined, and the obtained i-th prediction instruction is related to the i-th corresponding instruction, so that the i-th prediction instruction is a subordinate instruction of the i-1-th prediction instruction, that is, the i-th prediction instruction has more refined information, that is, as i starts to increase from 1, after the first large model instance determines the prediction instruction according to the demand information, other large model instances are used to further refine the prediction instruction, so that gradually refined prediction instructions are obtained through multiple large model instances. A single large model instance only needs to ensure its own instruction decision, and there is no need to pay attention to other large model instances, which greatly reduces the global dependence of each large model instance, so that each large model instance can better complete its own decision, improve the instruction determination accuracy of each large model instance, and then improve the determination accuracy of the demand matching instruction, and improve the overall stability and accuracy of the information processing process.

[0060] Specifically, when i≥2, the i-1th prediction instruction includes the i-1th instruction and the i-1th parameter, the i-th prediction instruction includes the i-th instruction and the i-th parameter, the i-1th instruction and the i-th instruction exist in the multi-layer initial instruction, and the i-th instruction is the subordinate instruction of the i-1th instruction, that is, the i-th large model instance has the ability to filter the i-th instruction from the i-th corresponding instruction after the adjustment of the i-th corresponding instruction, and determines the new i-th parameter according to the i-1th instruction and the i-1th parameter, so that in the processing from the i-1th instruction to the i-th instruction, the instruction is refined and the parameter is adjusted accordingly, realizing the call of the DSL language and the determination of the parameter. Since the determined prediction instruction is a DSL instruction, the platform can recognize the DSL instruction and execute it.

[0061] Specifically, the subordinate instructions of the i-1 prediction instruction can be determined by the instruction controller, which can be configured on the low-code platform, and the i-1 prediction instruction is input into the instruction controller. The instruction controller can determine the subordinate instructions of the i-1 prediction instruction based on the multiple layers of initial instructions, so as to determine the i-th largest model instance using the subordinate instructions of the i-1 prediction instruction. The subordinate instructions of the i-1 prediction instruction can include one instruction or multiple instructions.

[0062] The working process of several large model instances is described in detail below.

[0063] Based on the above description, it can be seen that the total number of layers of multi-layer initial instructions is n, the i-th largest model instance is determined according to the i-th corresponding instruction, the i-th largest model instance among multiple large model instances can be used to generate the i-th predicted instruction, i gradually increases from 1, and the maximum value is n.

[0064] During the working process of the large model instance, when i is 1, the first corresponding instruction is the initial instruction of the first layer in the multi-layer initial instructions, and the first prediction instruction is generated according to the demand information. The initial instruction of the first layer is used as the first corresponding instruction, and the large model is adjusted to obtain the first large model instance. The first large model instance is used as the general scheduling instance, and the demand information will be input into the first large model instance for analysis; the first large model instance generates the first prediction instruction according to the demand information, which is recorded as cmd-1, which corresponds to the initial instruction of the first layer, for example, it can be a generation-type instruction.

[0065] In the working process of the large model instance, when i≥2, the i-th pair of instructions is the subordinate instruction of the i-1th prediction instruction, and when i≥2, the i-th prediction instruction is generated according to the i-1th prediction instruction. Then, when i≥2, the i-th corresponding instruction can be determined according to the subordinate instruction of the i-1th prediction instruction cmd-i-1, and the large model is adjusted based on the i-th corresponding instruction to obtain the i-th large model instance llm-i, and the i-1th prediction instruction cmd-i-1 is input into the i-th large model instance llm-i, and the i-th large model instance llm-i generates the i-th prediction instruction cmd-i according to the i-1st prediction instruction, and then the low-code platform executes the i-th prediction instruction cmd-i to obtain the execution result Ri, and determines whether to call the lower-level large model to continue the instruction refinement according to the execution result Ri. If so, the current i value is added by 1 to obtain a new i value, and the i-th prediction instruction generated above is used as the i-1st prediction instruction, and the new i-th prediction instruction is generated again based on this. The training of the i-th model instance can be performed before it is determined that the i-th model instance needs to be called, or after it is determined that the i-th model instance needs to be called. If it is determined that the i-th model instance needs to be called, the i-th model instance will not be trained when it is not needed, thus saving computing resources.

[0066] Taking i as 2 as an example, the subordinate instruction of the first prediction instruction is determined as the second corresponding instruction according to the multi-layer initial instruction, and the large model is adjusted to obtain the second large model instance. If the first prediction instruction is a "generation type instruction", the second large model instance can be used as a generator. If the first prediction instruction is an "acquisition type instruction", the second large model instance can be used as an information acquirer. The following will be explained using the generator as an example; the first prediction instruction is input into the second large model instance, and the second large model instance can determine the second prediction instruction, which is a subordinate instruction of the first prediction instruction and has more detailed content, recorded as cmd-2, for example, it can be "generate data". Afterwards, the second prediction instruction can be executed to obtain the execution result R-2, and it is determined whether the instruction needs to be further refined according to the execution result R-2. For example, the third prediction instruction can be "generate dynamic data" and the fourth prediction instruction can be "generate data based on a database".

[0067] For ease of description, the i-th large model instance is taken as the subordinate large model instance of the i-1th large model instance. Relatively speaking, the i-1th large model instance is the superior large model instance of the i-th large model instance. In the process of generating the demand matching instruction, the superior large model instance can be called by the superior large model instance, and the superior large model instance is determined by the superior initial instruction. The superior large model instance can determine whether to call the superior large model instance based on the execution result of the prediction instruction generated by itself. The i-1th prediction instruction can be executed through the low-code platform to obtain the execution result, and the execution result can be returned to the i-1th large model instance for use by the i-1th large model instance.

[0068] Specifically, when i≥2, during the generation process of the i-th prediction instruction, the i-1th prediction instruction can be executed to obtain the execution result of the i-1th prediction instruction; if the execution result of the i-1th prediction instruction does not match the i-1th prediction instruction, it means that the execution result cannot meet the requirements of the i-1th prediction instruction, the i-1th prediction instruction is not detailed enough, and needs to be further refined, then the i-1th prediction instruction is input to the i-1th large model instance through the i-1th large model instance among the multiple large model instances; the i-1th prediction instruction is generated according to the i-1th prediction instruction through the i-th large model instance, so that the large model instances can be called step by step, and when the execution result matches the prediction instruction, the lower-level large model instance is not called, and the call is automatically carried out and stopped, thereby improving the instruction generation efficiency. The execution result of the i-1th prediction instruction and the matching determination of the i-1th prediction instruction can be determined by comparing the i-1th instruction and the i-1th parameter of the i-1th prediction instruction according to the information of the execution result, and the information of the execution result can include type, attribute, name, content, etc.

[0069] Through the above process, the large model instance can be called from top to bottom to form a progressive execution chain. In the actual operation process, the same large model instance may generate multiple prediction instructions, that is, the i-th prediction instruction or the i-1-th prediction instruction may include multiple ones. This is because one prediction instruction may not meet the demand, so multiple prediction instructions are used to work together to achieve the corresponding demand. Specifically, the execution order of multiple prediction instructions generated by the large model instance can be indicated by the thinking chain in the large model instance. The thinking chain is an improved prompt process for the large prediction model, and the purpose is to improve the performance of the large model in complex reasoning tasks. For example, for the demand information of "generating a bar chart to count the number of men and women in the merchant", it is necessary to implement it through the two prediction instructions of "generating data of male and female merchants" and "generating a bar chart". The execution order of these two prediction instructions is to execute "generating data of male and female merchants" first, and then execute "generating a bar chart". After the first large model instance determines these two prediction instructions according to the demand information, the two prediction instructions can be output in sequence.

[0070] Specifically, if there are multiple i-1 prediction instructions, the execution order of the i-1 prediction instructions can also be determined through the i-1 large model instance, and then the i-1 prediction instruction is executed according to the execution order of the i-1 prediction instruction to obtain the execution result of the i-1 prediction instruction (which can be one execution result or multiple execution results). If there is an i-1 target instruction among multiple i-1 prediction instructions that does not have a matching execution result, it means that the i-1 target instruction needs to be refined, and the i-1 large model instance in the multiple large model instances can be used according to the i-1 large model instance. The execution order of the i-th predicted instruction is input into the i-th model instance, so that when there are multiple i-1-th target instructions, the multiple i-1-th target instructions can be refined respectively through the multiple i-th model instances to obtain multiple i-th predicted instructions, so that the multiple i-th model instances can independently perform instruction refinement, and when there is no dependency between the multiple i-th model instances, they can be executed in parallel. Of course, when there is a dependency between the multiple i-th model instances, they can be executed in series, thereby expanding the instruction generation process and enabling it to meet more diverse scenarios.

[0071] If the requirements cannot be met after the separate refinement of the i-th large model instance, the obtained i-th prediction instruction can be further refined through their respective refinement links. The refinement process can be specifically as follows: if there are multiple i-1-th target instructions, after determining their respective i-th prediction instructions according to the obtained i-1-th target instructions through multiple i-th large model instances, the i-th prediction instructions can be executed separately to obtain the execution results of the i-th prediction instructions. If the execution results of the i-th target instructions in the multiple i-th prediction instructions do not match the i-th target instruction, it means that the i-th target instruction does not have a matching execution result. Then, by generating a target large model instance of the i-th target instruction, the i-th target instruction is input into the subordinate large model instance of the target large model. The target large model instance is determined by the target corresponding instruction in the i-th corresponding instruction, and the subordinate large model instance of the target large model instance is determined by the subordinate instruction of the target corresponding instruction. In this way, the call of the subordinate large model instance can be realized to construct different refinement links.

[0072] It should be noted that since the multiple i-1th target instructions generated by the same i-1th large model are respectively input into multiple i-th large model instances, these multiple i-th large model instances are all used as subordinate large model instances of the i-1th large model instance. When different i-th targets generated by different i-th large model instances are input into different i+1th large model instances, the i-th large model instance only inputs the i-th target instruction it generates into its own subordinate large model instance, and not into the subordinate large model instances of other i-th large model instances, thereby ensuring the independent refinement of each i-th target instruction. That is to say, only the upper-level large model instance and the lower-level large model instance belonging to the same link will transmit the prediction instructions and return the subsequent execution results. The upper-level large model instance and the lower-level large model instance are relative concepts determined according to the transmission path of the prediction instructions and the execution results. The upper-level large model instance transmits the prediction instructions to the lower-level large model instance, and the two have a continuous transmission path, which is equivalent to having a connection relationship and a superior-subordinate relationship. The two large model instances that do not have a transmission path for prediction instructions are equivalent to not having a connection relationship and not having a superior-subordinate relationship.

[0073] For example, refer to Figure 6 As shown, it is a schematic diagram of the working process of multiple large model instances provided in an embodiment of the present application, a first large model instance is used to generate two first prediction instructions of "generate data" and "generate UI components", and these two first prediction instructions are respectively input into two second large model instances, and these two second large model instances are both subordinate large model instances of the aforementioned first large model instance.

[0074] Among these two second largest model instances, one of the second largest model instances is used to generate the second prediction instruction of "generate dynamic data" according to the first prediction instruction of "generate data", and it can have a subordinate large model instance: the third largest model instance is used to generate the third prediction instruction of "generate data based on database", and the other second largest model instance is used to generate the second prediction instruction of "generate chart component" according to the first prediction instruction of "generate UI component", and it can have a subordinate large model instance: the third largest model instance is used to generate the third prediction instruction of "generate bar chart".

[0075] These two third largest model instances respectively refine different second prediction instructions, and both are connected to their superior large model instance (the second largest model instance connected to it), but not to the other second largest model instance. Therefore, the third largest model instance that generates "generates data based on database" and the second largest model instance that generates "generates chart-like components" do not have a superior-subordinate relationship, and the two are independent of each other. The third largest model instance that generates "generates a bar chart" and the second largest model instance that generates "generates dynamic data" do not have a superior-subordinate relationship, and the two are independent of each other.

[0076] If the i-1th predicted instruction includes a first instruction to be executed and a second instruction to be executed that can be executed simultaneously, the first instruction to be executed can be input into the i-th largest model instance corresponding to the first instruction to be executed, and the second instruction to be executed can be input into the i-th largest model instance corresponding to the second instruction to be executed, so that the two i-th largest model instances can execute instruction refinement in parallel.

[0077] If the i-1th predicted instruction includes a first instruction to be executed and a second instruction to be executed after the first instruction to be executed, if the first instruction to be executed and the second instruction to be executed do not have matching execution results, the first instruction to be executed is input into the i-th large model instance corresponding to the first instruction to be executed through the i-1th large model instance among multiple large model instances to obtain a refined instruction of the first instruction to be executed, wherein the execution result of the refined instruction of the first instruction to be executed matches the first instruction to be executed, and then the second instruction to be executed and the execution result of the refined instruction of the first instruction to be executed are sent to the i-th large model instance corresponding to the second instruction to be executed through the i-1th large model instance, wherein the execution result of the refined instruction of the first instruction to be executed is equivalent to the execution result of the first instruction to be executed, which can be used as a parameter of the second instruction to be executed. In this way, the first instruction to be executed can be refined through one execution link, and the second instruction to be executed can be refined through another execution link. Since the second instruction to be executed is after the first instruction to be executed, the second instruction to be executed can be refined after the first instruction to be executed is refined. In this way, the execution order between instructions at the same level is indicated by the thought chain, and the connection between the two execution links is realized, so that multiple refinement links are connected in series into a progressive execution link. In this way, using the multi-layer DSL as a transit medium, combined with the thought chain and the layered large model instance, the large language model can automatically control the low-code platform to generate instructions, thereby realizing application development.

[0078] For example, refer to Figure 6 As shown, the two predicted instructions "generate data" and "generate UI component" are respectively used as the first instruction to be executed and the second instruction to be executed. When it is determined that the detailed instruction of the first instruction to be executed, "generate data", is "generate data based on database", the execution result of the detailed instruction is determined, which can be used as a parameter of the second instruction to be executed, "generate UI component". In this way, after the detailed instruction of "generate UI component" is determined, the execution result of the detailed instruction matches the demand information, and the execution result can be used as response information to realize a progressive execution chain.

[0079] refer to Figure 7 As shown, it is a schematic diagram of the working process of another multiple large model instances provided in the embodiment of the present application. Taking the demand information of "generating a bar chart of the number of male and female merchants" as an example, the generation process of the demand matching instruction is as follows:

[0080] 1. The first large model instance (i.e., the total scheduling instance of the large model) is obtained by training through the initial instruction of the first layer as the first corresponding instruction. The initial instruction of the first layer includes generation class instructions and acquisition class instructions, which are expressed as generation (xxx) and acquisition (xxx), and xxx corresponds to the parameter of the instruction;

[0081] 2. Input demand information, the demand information is "generate a bar chart to count the number of men and women in a merchant", the first large model instance generates the first prediction instruction according to the demand information, the first prediction instruction includes the first instruction and the second instruction, the first instruction is "generate data", and the second instruction is "generate UI component";

[0082] 3. The initial instructions of the second layer are used as the second corresponding instructions to train a second large model instance (i.e., the first generator). The initial instructions of the second layer include generate UI (xxx), generate data (xxx), and other generation instructions. xxx corresponds to the parameters of the instruction. The second large model instance does not need to pay attention to the context of other large models in the upper layer, and can make decisions independently according to the needs.

[0083] 4. The first large model instance inputs the first instruction in the first prediction instruction to the second large model instance (i.e., the first generator), and the second large model instance determines the second prediction instruction according to the first instruction;

[0084] 5. The second largest model instance (i.e., the first generator) outputs the second prediction instruction, which is used as the (i-1)th prediction instruction, and starts step 6;

[0085] 6. Input the i-1th predicted instruction into the instruction controller;

[0086] 7. Determine, by executing the controller based on the multi-layer initial instructions, the lower-level instruction of the i-1th predicted instruction as the i-th corresponding instruction;

[0087] 8. Determine the i-th largest model instance based on the i-th corresponding instruction;

[0088] 9. The i-1th prediction instruction is input into the i-th largest model instance to obtain the i-th prediction instruction;

[0089] 10. Output the i-th prediction instruction so that the i-th prediction instruction can be executed through the low-code platform to obtain the execution result Ri;

[0090] 11. Determine whether further refinement is required based on the execution result Ri of the i-th large model instance, that is, determine whether to call the lower-level large model instance. If so, execute 12, otherwise execute 13;

[0091] 12. i+1 is used as the new i, and the i-th prediction instruction is used as the new i-1-th prediction instruction, and the process returns to step 6;

[0092] 13. Determine that the i-th prediction instruction is a refinement instruction of the first prediction instruction corresponding to the current refinement, and the execution result of the i-th prediction instruction is used as the execution result of the first prediction instruction; the refinement instruction of the first instruction may be "generate data based on database", and the parameter is "the number of men and women in the merchant", and the refinement instruction of the second instruction may be "generate a bar chart", and the parameter is "the data of the number of men and women in the merchant";

[0093] 14. Return the execution result of the i-th prediction instruction to the upper large model instance (e.g., the first generator), and the return method may be layer by layer or to the large model instance of the preset layer;

[0094] 15. The first generator returns the execution result of the i-th prediction instruction to the first large model instance as the execution result of the first instruction in the first prediction instruction;

[0095] 16. Using the initial instructions of the second layer as the second corresponding instructions, another second large model instance (i.e., the second generator) is trained. The initial instructions of the second layer include generate UI (xxx), generate data (xxx) and other generation instructions. xxx corresponds to the parameters of the instructions.

[0096] 17. The first large model instance performs refinement on the second instruction, inputs the second instruction in the first prediction instruction and the execution result of the first instruction to the second large model instance (the second generator), and the second large model instance determines the second prediction instruction according to the second instruction, and the execution result of the first instruction can be used as a parameter of the second instruction;

[0097] 18. The second large model instance (second generator) outputs the second prediction instruction and uses it as the (i-1)th prediction instruction, and starts step 4 until a new (i)th prediction instruction is obtained as the refinement instruction of the second instruction, and the execution result of the refinement instruction of the second instruction is used as the execution result of the second instruction;

[0098] 19. Return the execution result of the refined instruction of the second instruction to the upper-level large model instance (e.g., the second generator);

[0099] 20. The second generator returns the execution result of the new i-th prediction instruction to the first large model instance as the execution result of the second instruction in the first prediction instruction;

[0100] 21. When the first large model instance determines that the execution result of the second instruction matches the demand information, it returns the execution result of the second instruction as the response information of the demand information. If it does not meet the demand information, the first prediction instruction can be re-determined.

[0101] The detailed instructions of the first instruction include the instruction of "generate data based on database" and the parameter of "the number of men and women in the merchant". After execution by the platform, this detailed instruction returns the execution result of "the data of the number of men and women in the merchant"; the above-mentioned execution result can be used as the parameter of the second instruction. The detailed instructions of the second instruction include the instruction of "generate a bar chart" and the parameter of "the data of the number of men and women in the merchant". The low-code platform generates a bar chart according to the detailed instructions of the second instruction, and returns "the bar chart is generated successfully" to the large model instance. The large model instance returns the execution result of the detailed instructions of the second instruction to the large model general scheduling. The large model general scheduling judgment process ends and the generation task is successfully completed.

[0102] When the large model instances are called layer by layer in sequence, each large model instance executes the instruction refinement in series, and the generated prediction instruction is refined layer by layer. Then the final prediction instruction is generated by the large model instance at the bottom layer. The execution result of the prediction instruction generated by the large model instance at the bottom layer can meet the needs of its own prediction instruction and the needs of the prediction instruction generated by the large model instance at the upper layer. Therefore, the large model instance at the lower layer can return the execution result of the prediction instruction generated by it to the large model instance at the upper layer. The execution result can be transmitted to the first large model instance by returning layer by layer, or by transmitting to the preset large model instance. By calling the large model instance downward layer by layer, and returning the execution result upward layer by layer, the large model instance at the upper layer controls the end timing of the process, realizing the automatic determination of the large model instance and the automatic generation of instructions.

[0103] Specifically, if i≥2, and the execution result of the i-th prediction instruction matches the first prediction instruction, the i-th prediction instruction is used as the refinement instruction of the first prediction instruction. Accordingly, the demand matching instruction can be determined according to the refinement instruction of the first prediction instruction. Generally speaking, when i>1, since the first prediction instruction is determined according to the demand information, if the execution result of the i-th prediction instruction matches the demand information, it means that the execution result of the i-th prediction instruction matches the first prediction instruction, and it is a refinement instruction of the first prediction instruction, that is, the i-th prediction instruction not only meets the demand expressed by the first prediction instruction so that the process no longer calls the lower-level large model instance, but also meets the demand expressed by the demand information so that the process ends the determination process of the prediction instruction. The first prediction instruction can be one or more, and the refinement instruction of a first prediction instruction can be one or more, thereby realizing the refinement of the first prediction instruction and realizing the determination from the prediction instruction to the demand matching instruction. For example, the detailed instructions of the two prediction instructions "generate data" and "generate UI components" are "generate data based on database" and "generate bar chart", and both are demand matching instructions, that is, executing these two instructions can determine the response information of the demand information.

[0104] In the embodiment of the present application, the execution result of the prediction instruction can be returned upward layer by layer, and the determination of the refined instruction of the prediction instruction can be realized by returning the execution result of the large model instance upward layer by layer. Specifically, if i≥2 and the execution result of the ith prediction instruction matches the ith prediction instruction, the ith large model instance returns the execution result of the ith prediction instruction to the upper-level large model instance of the ith large model instance, so that when the upper-level large model instance of the ith large model instance is not the first large model instance, the upper-level large model instance returns the execution result of the ith prediction instruction to the upper-level large model instance of the upper-level large model instance when determining that the execution result of the ith prediction instruction matches the prediction instruction generated by the upper-level large model instance, until the execution result of the ith prediction instruction is returned to the first large model instance. The upper-level large model instance of the upper-level large model instance of the ith large model instance is determined by the preset instruction in the i-2 corresponding instruction, the upper-level large model instance of the ith large model instance is determined by the lower-level instruction of the preset instruction, and the ith large model instance is determined by the lower-level instruction of the lower-level instruction of the preset instruction.

[0105] After the first large model instance obtains the ith prediction instruction, if it is determined that the execution result of the ith prediction instruction matches the demand information, the execution result of the ith prediction instruction can be determined as the response information; in addition, if the first large model instance determines that the execution result of the ith prediction instruction matches the first prediction instruction, the ith prediction instruction will be used as a refined instruction of the ith prediction instruction, and the refined instruction matches the prediction instruction determined by the upper-level large model instance, and can meet the demand expressed by the first prediction instruction, that is, the accurate refinement of the first prediction instruction is achieved.

[0106] In the specific implementation, there is still a situation where the returned execution result does not match the prediction instruction generated by the large model instance. Specifically, if i>j≥1, the execution result of the ith prediction instruction is returned to the jth large model instance, indicating that the ith prediction instruction matches the prediction instructions of the lower-level large model instance of the jth large model instance. If the jth large model instance determines that the execution result of the ith prediction instruction does not match the jth prediction instruction, it means that even if the lower-level large model instance executes the correct refinement process, it cannot meet the needs of the jth prediction instruction. At this time, it can be considered that the jth prediction instruction is determined incorrectly, and a new jth prediction instruction can be determined based on the jth corresponding instruction to determine other instructions other than the jth prediction instruction, that is, the jth prediction instruction is re-determined from the jth corresponding instruction, and then the new jth prediction instruction is determined, thereby realizing the correction of the jth prediction instruction. When the jth prediction instruction includes the jth instruction and the jth parameter, the process can re-determine the jth instruction and the jth parameter.

[0107] After the response information is determined, the response information can be displayed through the terminal device so that the user can obtain the response information to the demand information. Of course, the demand matching instruction and response information can also be displayed through the terminal device, and the execution process prompt information can also be displayed after the demand matching instruction is executed to reasonably respond to user needs.

[0108] refer to Figure 8 As shown, a schematic diagram of a graphical user interface of another low-code platform provided in an embodiment of the present application, after determining the demand matching instruction, the demand matching instruction can be displayed in the first platform bubble 141 in the chat box 10, and the execution process prompt information can be displayed in the second platform bubble 142 in the chat box 10. The demand matching instruction includes: the first instruction: "data = addData('the number of male and female merchants')", and the second instruction: (createComponent("bar chart", data)), the execution process prompt information includes: "the instruction is executed successfully, and the process ends", and the response information can be displayed in the result display interface of the low-code platform.

[0109] Based on an information processing method provided in an embodiment of the present application, an information processing device is also provided in an embodiment of the present application, referring to Fig. 9 , which is a structural block diagram of an information processing device provided in an embodiment of the present application, the information processing device 1300 includes:

[0110] The information acquisition unit 1301 is used to acquire demand information;

[0111] The information processing unit 1302 is configured to determine response information according to the demand information through multiple large model instances; in the process of determining the response information, generate an i-th prediction instruction through an i-th large model instance among the multiple large model instances, and if the execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer;

[0112] Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

[0113] Optionally, when i≥2, the information processing unit 1302 includes:

[0114] an execution unit, configured to execute the i-1th prediction instruction to obtain an execution result of the i-1th prediction instruction;

[0115] an input unit, configured to input the i-1th prediction instruction to the i-th large model instance through the i-1th large model instance among the multiple large model instances if the execution result of the i-1th prediction instruction does not match the i-1th prediction instruction;

[0116] A prediction unit is used to generate an i-th prediction instruction according to the i-1-th prediction instruction through the i-th largest model instance.

[0117] Optionally, if the number of the i-1th prediction instructions is multiple, the device further includes:

[0118] an execution order determination unit, configured to determine the execution order of the plurality of i-1th predicted instructions by using the i-1th large model instance;

[0119] The execution unit is specifically used for: executing the plurality of the i-1th prediction instructions according to the execution order to obtain the execution result of the i-1th prediction instruction;

[0120] The input unit is specifically used for:

[0121] If there is an i-1th target instruction among the multiple i-1th prediction instructions that does not have a matching execution result, then the corresponding i-1th target instruction is input into the i-th large model instance according to the execution order through the i-1th large model instance among the multiple large model instances.

[0122] Optionally, if there are multiple i-1th target instructions, the device further includes:

[0123] An instruction prediction unit, configured to determine respective i-th predicted instructions according to the obtained i-1-th target instructions respectively through the plurality of i-th large model instances;

[0124] an instruction execution unit, configured to execute the plurality of the i-th prediction instructions respectively to obtain execution results of the i-th prediction instructions;

[0125] An instruction input unit is used to input the i-th target instruction to the subordinate large model instance of the target large model instance by generating a target large model instance of the i-th target instruction if there is an i-th target instruction among multiple i-th prediction instructions and the execution result of the i-th target instruction does not match the i-th target instruction, the target large model instance is determined by the target corresponding instruction in the i-th corresponding instruction, and the subordinate large model instance of the target large model instance is determined by the subordinate instruction of the target corresponding instruction.

[0126] Optionally, if the (i-1)th predicted instruction includes a first instruction to be executed and a second instruction to be executed after the first instruction to be executed, the instruction input unit includes:

[0127] a first input unit, configured to input the first instruction to be executed to the i-th large model instance corresponding to the first instruction to be executed through the i-1th large model instance among the multiple large model instances, if the first instruction to be executed and the second instruction to be executed do not have matching execution results, so as to obtain a refined instruction of the first instruction to be executed, and the execution result of the refined instruction of the first instruction to be executed matches the first instruction to be executed;

[0128] The second input unit is used to input the second instruction to be executed and the execution result of the refined instruction of the first instruction to be executed to the i-th largest model instance corresponding to the second instruction to be executed through the i-1th largest model instance.

[0129] Optionally, the device further comprises:

[0130] a refinement instruction determination unit, configured to use the i-th prediction instruction as a refinement instruction of the first prediction instruction if i≥2 and the execution result of the i-th prediction instruction matches the first prediction instruction;

[0131] The demand matching instruction determining unit is used to determine the demand matching instruction according to the refined instruction of the first prediction instruction.

[0132] Optionally, the device further comprises:

[0133] An execution result returning unit, used for returning the execution result of the i-th prediction instruction to the upper-level large model instance of the i-th large model instance if i≥2 and the execution result of the i-th prediction instruction matches the i-th prediction instruction, so that when the upper-level large model instance is not the first large model instance, the upper-level large model instance returns the execution result of the i-th prediction instruction to the upper-level large model instance of the upper-level large model instance when determining that the execution result of the i-th prediction instruction matches the prediction instruction generated by the upper-level large model instance, until the execution result of the i-th prediction instruction is returned to the first large model instance; the upper-level large model instance of the upper-level large model instance is determined by the preset instruction in the i-2 corresponding instruction, the upper-level large model instance is determined by the lower-level instruction of the preset instruction, and the i-th large model instance is determined by the lower-level instruction of the lower-level instruction of the preset instruction.

[0134] Optionally, the device further comprises:

[0135] An instruction correction unit is used to, if i>j≥1, return the execution result of the i-th prediction instruction to the j-th model instance, and the j-th model instance determines that the execution result of the i-th prediction instruction does not match the j-th prediction instruction, then determine a new j-th prediction instruction based on the j-th corresponding instruction to determine other instructions other than the j-th prediction instruction.

[0136] Optionally, the device further comprises:

[0137] A display unit is used to display the response information through a terminal device; or to display the demand matching instruction and the response information through a terminal device.

[0138] Optionally, the device further comprises:

[0139] The model training unit is used to adjust the model to be optimized according to the i-th corresponding instruction to obtain the i-th largest model instance, and the model to be optimized is obtained by pre-training the initial model.

[0140] Optionally, when i≥2, the i-1th prediction instruction includes the i-1th instruction and the i-1th parameter, the i-th prediction instruction includes the i-th instruction and the i-th parameter, the i-1th instruction and the i-th instruction exist in the multiple layers of initial instructions, and the i-th instruction is a subordinate instruction of the i-1th instruction.

[0141] It can be seen from the above technical solution that a multi-layer initial instruction of a domain-specific language is provided, and the i-th layer instruction in the multi-layer initial instruction includes the i-th corresponding instruction. The initial instruction at the i-th layer has detailed information relative to the initial instruction at the i-1th layer, and i is a positive integer, that is, the multi-layer initial instruction is refined layer by layer from top to bottom, so that the operation content of the multi-layer initial instruction is clarified layer by layer. The i-th large model instance is determined according to the i-th corresponding instruction. The i-th corresponding instruction is the instruction of the i-th layer in the multi-layer initial instruction, and is a subordinate instruction of the i-1th prediction instruction. In this way, the semantics of the multi-layer initial instruction is integrated into the large model instance, so that the response process to the demand information can be combined with the advantages and capabilities of the specific domain language, which is conducive to the expansion of functions. Different large model instances can use different initial instructions, and the initial instructions of different layers are used to determine different large model instances. The initial instructions of each layer are independent of each other and there is no problem of mutual interference, which is conducive to improving the response accuracy of each large model instance.

[0142] After obtaining the demand information, the response information can be determined according to the demand information through multiple large model instances. In the process of determining the response information, the i-th prediction instruction can be generated through the i-th large model instance among the multiple large model instances. If the execution result of the i-th prediction instruction matches the demand information, the execution result of the i-th prediction instruction is used as the response information. i is a positive integer, that is, multiple large model instances can be used to realize the response to the demand information. Multiple large model instances work together to improve the accuracy and efficiency of demand response. Among them, the first prediction instruction is generated according to the demand information. When i≥2, the i-th prediction instruction is generated according to the i-1 prediction instruction. The generated i-th prediction instruction is related to the i-th corresponding instruction, so that the i-th prediction instruction is the subordinate instruction of the i-1 prediction instruction. That is, after the first large model instance determines the prediction instruction according to the demand information, other large model instances are used to further refine the prediction instruction, so that gradually refined prediction instructions are obtained through multiple large model instances. A single large model instance only needs to ensure its own instruction decision, which reduces the global dependence of each large model instance, improves the instruction determination accuracy of each large model instance, and then improves the determination accuracy of the response information, and improves the overall stability and accuracy of the information processing process.

[0143] The embodiment of the present application further provides a computer device, which is the computer device described above and may include a terminal device or a server, and the information processing device described above may be configured in the computer device. The computer device is described below in conjunction with the accompanying drawings.

[0144] If the computer device is a terminal device, see Fig.10 As shown, the embodiment of the present application provides a terminal device, taking the terminal device as a mobile phone as an example:

[0145] Fig.10 FIG. 1 is a block diagram showing a partial structure of a mobile phone related to a terminal device provided in an embodiment of the present application. Fig.10 The mobile phone includes: a radio frequency (RF) circuit 1410, a memory 1420, an input unit 1430, a display unit 1440, a sensor 1450, an audio circuit 1460, a wireless fidelity (WiFi) module 1470, a processor 1480, and a power supply 1490. Those skilled in the art can understand that Fig.10 The mobile phone structure shown in the figure does not constitute a limitation on the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0146] Combine the following Fig.10 A detailed introduction to the various components of the mobile phone:

[0147] The RF circuit 1410 may be used for receiving and sending signals during information transmission or communication or during a call. In particular, after receiving downlink information from a base station, the information is sent to the processor 1480 for processing. In addition, the uplink data is sent to the base station.

[0148] The memory 1420 can be used to store software programs and modules. The processor 1480 executes various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 1420. The memory 1420 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 1420 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0149] The input unit 1430 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the mobile phone. Specifically, the input unit 1430 may include a touch panel 1431 and other input devices 1432 .

[0150] The display unit 1440 may be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 1440 may include a display panel 1441.

[0151] The mobile phone may also include at least one sensor 1450, such as a light sensor, a motion sensor, and other sensors.

[0152] The audio circuit 1460 , the speaker 1461 , and the microphone 1462 can provide an audio interface between the user and the mobile phone.

[0153] WiFi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse web pages and access streaming media through the WiFi module 1470. It provides users with wireless broadband Internet access.

[0154] Processor 1480 is the control center of the mobile phone. It uses various interfaces and lines to connect various parts of the entire mobile phone. It executes various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in memory 1420 and calling data stored in memory 1420.

[0155] The mobile phone also includes a power source 1490 (such as a battery) for supplying power to various components.

[0156] In this embodiment, the processor 1480 included in the terminal device also has the following functions:

[0157] Obtain demand information;

[0158] Determine response information according to the demand information through multiple large model instances; in the process of determining the response information, generate an i-th prediction instruction through the i-th large model instance among the multiple large model instances, and if the execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer;

[0159] Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

[0160] If the computer device is a server, the present application embodiment also provides a server, see Fig.11 As shown, Fig.11 The structural diagram of the server 1500 provided in the embodiment of the present application, the server 1500 may have relatively large differences due to different configurations or performances, and may include one or more processors 1522, such as a central processing unit (CPU), a memory 1532, and one or more storage media 1530 (e.g., one or more mass storage devices) storing application programs 1542 or data 1544. Among them, the memory 1532 and the storage medium 1530 can be short-term storage or permanent storage. The program stored in the storage medium 1530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the server. Furthermore, the processor 1522 can be configured to communicate with the storage medium 1530 and execute a series of instruction operations in the storage medium 1530 on the server 1500.

[0161] The server 1500 may also include one or more power supplies 1526, one or more wired or wireless network interfaces 1550, one or more input and output interfaces 1558, and / or one or more operating systems 1541, such as Windows Server 2003 or Windows Server 2008. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TMetc.

[0162] The steps performed by the server in the above embodiment can be based on Fig.11 The server structure shown.

[0163] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method provided in the above embodiment.

[0164] The embodiments of the present application also provide a computer program product including a computer program, which, when executed on a computer device, enables the computer device to execute the method provided in the above embodiments.

[0165] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to the computer program, and the aforementioned computer program can be stored in a computer-readable storage medium. When the computer program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned computer-readable storage medium can be at least one of the following media: read-only memory (English: Read-only Memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc. Various media that can store computer programs.

[0166] It should be noted that each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0167] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a technician familiar with the technical field within the technical scope disclosed in the present application should be included in the protection scope of the present application. Moreover, based on the implementation methods provided in the above aspects, the present application can also be further combined to provide more implementation methods. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An information processing method, characterized in that: The method comprises: Obtain demand information; Determine response information according to the demand information through multiple large model instances; in the process of determining the response information, generate an i-th prediction instruction through the i-th large model instance among the multiple large model instances, and if the execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer; Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

2. The method according to claim 1, characterized in that When i≥2, generating the i-th prediction instruction by using the i-th large model instance among the multiple large model instances includes: Executing the i-1th prediction instruction to obtain an execution result of the i-1th prediction instruction; If the execution result of the i-1th prediction instruction does not match the i-1th prediction instruction, inputting the i-1th prediction instruction to the i-th large model instance through the i-1th large model instance among the multiple large model instances; An i-th prediction instruction is generated according to the i-1-th prediction instruction through the i-th largest model instance.

3. The method according to claim 2, characterized in that If the number of the i-1th prediction instructions is multiple, the method further includes: Determining the execution order of the plurality of i-1th predicted instructions by using the i-1th large model instance; The executing the i-1th prediction instruction to obtain the execution result of the i-1th prediction instruction comprises: executing a plurality of the i-1th prediction instructions according to the execution order to obtain the execution result of the i-1th prediction instruction; If the execution result of the i-1th prediction instruction does not match the i-1th prediction instruction, inputting the i-1th prediction instruction to the i-th large model instance through the i-1th large model instance among the multiple large model instances comprises: If there is an i-1th target instruction among the multiple i-1th prediction instructions that does not have a matching execution result, then the corresponding i-1th target instruction is input into the i-th large model instance according to the execution order through the i-1th large model instance among the multiple large model instances.

4. The method according to claim 3, characterized in that If there are multiple i-1th target instructions, the method further includes: Determine respective i-th predicted instructions according to the obtained i-1-th target instructions by using the plurality of i-th large model instances; Executing the plurality of i-th prediction instructions respectively to obtain execution results of the i-th prediction instructions; If there is an i-th target instruction among multiple i-th prediction instructions, and the execution result of the i-th target instruction does not match the i-th target instruction, the i-th target instruction is input into the subordinate large model instance of the target large model instance by generating a target large model instance of the i-th target instruction, and the target large model instance is determined by the target corresponding instruction in the i-th corresponding instruction, and the subordinate large model instance of the target large model instance is determined by the subordinate instruction of the target corresponding instruction.

5. The method according to claim 3, characterized in that: If the i-1th predicted instruction includes a first instruction to be executed and a second instruction to be executed after the first instruction to be executed, and there is an i-1th target instruction in the multiple i-1th predicted instructions that does not have a matching execution result, then inputting the corresponding i-1th predicted instruction to the multiple i-th large model instances respectively according to the execution order through the i-1th large model instance among the multiple large model instances, including: If the first instruction to be executed and the second instruction to be executed do not have matching execution results, inputting the first instruction to be executed into the i-th large model instance corresponding to the first instruction to be executed through the i-1th large model instance among the multiple large model instances to obtain a refined instruction of the first instruction to be executed, and the execution result of the refined instruction of the first instruction to be executed matches the first instruction to be executed; The second instruction to be executed and the execution result of the refined instruction of the first instruction to be executed are input into the i-th largest model instance corresponding to the second instruction to be executed through the i-1th largest model instance.

6. The method according to claim 2, characterized in that The method further comprises: If i≥2 and the execution result of the i-th prediction instruction matches the first prediction instruction, the i-th prediction instruction is used as a refinement instruction of the first prediction instruction; Demand matching instructions are determined according to the refined instructions of the first prediction instructions.

7. The method according to claim 6, characterized in that The method further comprises: Displaying the response information through a terminal device; Or, the demand matching instruction and the response information are displayed through a terminal device.

8. The method according to claim 2, characterized in that: The method further comprises: If i≥2 and the execution result of the i-th prediction instruction matches the i-th prediction instruction, the i-th large model instance returns the execution result of the i-th prediction instruction to the upper-level large model instance of the i-th large model instance, so that when the upper-level large model instance is not the first large model instance, the upper-level large model instance returns the execution result of the i-th prediction instruction to the upper-level large model instance of the upper-level large model instance when determining that the execution result of the i-th prediction instruction matches the prediction instruction generated by the upper-level large model instance, until the execution result of the i-th prediction instruction is returned to the first large model instance; the upper-level large model instance of the upper-level large model instance is determined by the preset instruction in the i-2 corresponding instruction, the upper-level large model instance is determined by the lower-level instruction of the preset instruction, and the i-th large model instance is determined by the lower-level instruction of the lower-level instruction of the preset instruction.

9. The method according to claim 8, characterized in that The method further comprises: If i>j≥1, the execution result of the i-th prediction instruction is returned to the j-th model instance, and the j-th model instance determines that the execution result of the i-th prediction instruction does not match the j-th prediction instruction, then a new j-th prediction instruction is determined based on the j-th corresponding instruction to determine other instructions other than the j-th prediction instruction.

10. The method according to any one of claims 1 to 9, characterized in that: The method further comprises: The model to be optimized is adjusted according to the i-th corresponding instruction to obtain the i-th largest model instance, and the model to be optimized is obtained by pre-training the initial model.

11. The method according to any one of claims 1 to 9, characterized in that: When i≥2, the i-1th prediction instruction includes the i-1th instruction and the i-1th parameter, the i-th prediction instruction includes the i-th instruction and the i-th parameter, the i-1th instruction and the i-th instruction exist in the multi-layer initial instructions, and the i-th instruction is a subordinate instruction of the i-1th instruction.

12. An information processing device, characterized in that: The device comprises: An information acquisition unit, used for acquiring demand information; an information processing unit, configured to determine response information according to the demand information through a plurality of large model instances; in the process of determining the response information, generate an i-th prediction instruction through an i-th large model instance among the plurality of large model instances, and if an execution result of the i-th prediction instruction matches the demand information, use the execution result of the i-th prediction instruction as the response information, where i is a positive integer; Among them, the i-th large model instance is determined according to the i-th corresponding instruction, the i-th corresponding instruction is located at the i-th layer in the multi-layer initial instructions of the domain-specific language, the initial instructions located at the i-th layer have detailed information relative to the initial instructions located at the i-1-th layer, and when i≥2, the i-th corresponding instruction is a subordinate instruction of the i-1-th prediction instruction; the first prediction instruction is generated according to the demand information, and when i≥2, the i-th prediction instruction is generated according to the i-1-th prediction instruction.

13. A computer device, characterized in that: The computer device comprises a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is used to execute the information processing method described in any one of claims 1 to 11 according to the instructions in the computer program.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the information processing method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program, characterized in that When it is run on a computer device, the computer device is enabled to execute the information processing method described in any one of claims 1 to 11.

Images