Task processing system and method based on large model and computer program product

Abstract

The invention provides a task processing system and method based on a large model, and a computer program product, and relates to the technical field of artificial intelligence. The system comprises a task initiator, a task simulator and a large model, wherein the task initiator submits a task request; the task simulator analyzes task description information in the task request, constructs a directed acyclic graph based on key elements and a logic structure of a task, generates a task set containing a plurality of subtasks and an ordered task queue according to the directed acyclic graph, then calls a large model to obtain a processing result of each subtask in the task set, and sends the processing result to the task simulator; according to the output sequence of the processing results of the subtasks in the ordered task queue, the processing results of the subtasks are returned to the task initiator; therefore, each sub-task can be processed by the large model at one time, the processing result of each sub-task does not exceed the output limit of the large model, and the problem that the scale of the output token is limited when the large model processes a complex single task is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a task processing system and method based on a large model, and a computer program product. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (LLMs) have made remarkable achievements in the field of natural language processing (NLP). These models have demonstrated powerful capabilities in a variety of application scenarios such as text generation, machine translation, and question-answering systems. However, despite the excellent performance of LLMs in many tasks, they still face an important technical bottleneck - the limitation of the output token scale. The token here can be translated into a word unit, which is a basic concept in natural language processing. It is to divide text data into smaller units, which can be words, characters, or phrases. For example, the text data "I like artificial intelligence" can be divided into three words, namely "I", "like", and "artificial intelligence". In addition, the text data "I like artificial intelligence" can also be divided into four words, namely "I", "like", "artificial", and "intelligent", depending on the specific word segmentation method and application scenario.

[0003] In order to ensure operational efficiency and stability, most LLMs set a fixed upper limit for output tokens, which results in the model being unable to directly provide complete and coherent results when processing tasks that require a large amount of continuous text output, such as writing long reports and generating complex documents. In related technologies, attempts are made to increase the scale of output tokens by adjusting model parameters, but this can lead to a decrease in model performance, such as longer response time and reduced output quality; or, users need to decompose the task and submit it multiple times, which increases the complexity and time cost of use, and users are prone to errors when manually integrating multiple parts of the model output, affecting the quality and consistency of the final output.

[0004] Therefore, how to effectively solve the problem of token scale limitation of large language models without sacrificing performance is of great significance for improving the comprehensiveness and completeness of generated content, improving user experience, and expanding the application scope of large language models. Summary of the invention

[0005] In view of the above problems, the present application is proposed to provide a large model-based task processing system and method, computer program product, and storage medium that overcome the above problems or at least partially solve the above problems. The technical solution is as follows: In a first aspect, a task processing system based on a large model is provided, the system comprising a task initiator, a task simulator and a large model; The task initiator calls the interface of the task simulator to submit a task request to the task simulator; The task simulator receives the task request, parses the task description information in the task request, and identifies the key elements and logical structure of the task; The task simulator constructs a directed acyclic graph based on the key elements and logical structure of the task, wherein the directed acyclic graph includes a plurality of subtask nodes with a logical order, and the plurality of subtask nodes are used to store the task contents and constraint information of the plurality of subtasks; a task set including a plurality of subtasks is generated according to the directed acyclic graph; and an ordered task queue that specifies the output order of the processing results of each subtask is generated according to the directed acyclic graph; wherein each subtask in the task set includes the task content and constraint information, and each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model; The task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model; The big model generates the processing results of each subtask according to the task content and constraint information of each subtask, and returns them to the task simulator; The task simulator receives the processing results of each subtask, and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue.

[0006] In a possible implementation, the constraint information includes one or more of format requirements, timeline historical background, content depth requirements, and content breadth requirements.

[0007] In a possible implementation, a subtask in a task set may be called a current subtask; The constraint information also includes summary information of other subtasks associated with the current subtask, and the summary information of other subtasks is obtained through the task contents of other subtasks.

[0008] In a possible implementation, a subtask node among the multiple subtask nodes represents a subtask, a first subnode of the subtask node stores the task content of the subtask, and a second subnode of the subtask node stores the constraint information of the subtask.

[0009] In a possible implementation, the task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model, including: The task simulator calls the interface of the large model, randomly selects the task content and constraint information of one or more subtasks from the task set according to the set period and sends them to the large model until the task content and constraint information of all subtasks in the task set are sent to the large model.

[0010] In a possible implementation, if the output order of the processing results of each subtask in the ordered task queue is the processing result of the first subtask to the processing result of the Nth subtask, N is a positive integer; The task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model, including: The task simulator calls the interface of the large model, and based on the output order of the processing results of each subtask in the ordered task queue, sends the task content and constraint information of the first subtask in the task set, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask to the large model in sequence.

[0011] In a possible implementation, the task simulator receives the processing results of each subtask and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue, including: The task simulator receives the processing results of each subtask, and according to the output order of the processing results of each subtask in the ordered task queue, if the processing result of the current subtask is received, and the processing result of the subtask before the current subtask has not been received, it waits to receive the processing result of the subtask before the current subtask. When the processing result of the subtask before the current subtask is received, the processing results of the subtask before the current subtask and the processing result of the current subtask are returned to the task initiator in sequence.

[0012] In a second aspect, a large model-based task processing method is provided, which is applied to a task simulator, and the method comprises: Receive a task request submitted by a task initiator by calling an interface of a task simulator; Parsing the task description information in the task request to identify the key elements and logical structure of the task; Based on the key elements and logical structure of the task, a directed acyclic graph is constructed, wherein the directed acyclic graph includes a plurality of subtask nodes with a logical order, and the plurality of subtask nodes are used to store the task content and constraint information of the plurality of subtasks; according to the directed acyclic graph, a task set including the plurality of subtasks is generated; and according to the directed acyclic graph, an ordered task queue is generated that specifies the output order of the processing results of each subtask; wherein each subtask in the task set includes the task content and constraint information, and each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model; Call the interface of the big model to send the task content and constraint information of each subtask in the task set to the big model; Receive the processing results of each subtask returned by the large model, and return the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue.

[0013] In a possible implementation, the constraint information includes one or more of format requirements, timeline historical background, content depth requirements, and content breadth requirements.

[0014] In a possible implementation, a subtask in a task set may be called a current subtask; The constraint information also includes summary information of other subtasks associated with the current subtask, and the summary information of other subtasks is obtained through the task contents of other subtasks.

[0015] In a possible implementation, a subtask node among the multiple subtask nodes represents a subtask, a first subnode of the subtask node stores the task content of the subtask, and a second subnode of the subtask node stores the constraint information of the subtask.

[0016] In a possible implementation, the interface of the big model is called to send the task content and constraint information of each subtask in the task set to the big model, including: The interface of the large model is called, and the task content and constraint information of one or more subtasks are randomly selected from the task set according to the set period and sent to the large model until the task content and constraint information of all subtasks in the task set are sent to the large model.

[0017] In a possible implementation, if the output order of the processing results of each subtask in the ordered task queue is the processing result of the first subtask to the processing result of the Nth subtask, N is a positive integer; Call the interface of the big model to send the task content and constraint information of each subtask in the task set to the big model, including: Call the interface of the big model, and based on the output order of the processing results of each subtask in the ordered task queue, send the task content and constraint information of the first subtask, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask in the task set to the big model in sequence.

[0018] In a possible implementation, receiving the processing results of each subtask returned by the large model, and returning the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue, includes: Receive the processing results of each subtask returned by the large model, and according to the output order of the processing results of each subtask in the ordered task queue, if the processing result of the current subtask is received, and the processing result of the subtask before the current subtask has not been received, then wait to receive the processing result of the subtask before the current subtask. When the processing result of the subtask before the current subtask is received, return the processing results of the subtask before the current subtask and the processing result of the current subtask to the task initiator in sequence.

[0019] According to a third aspect, a computer program product is provided, comprising a computer program, wherein the computer program is configured to execute any one of the large model-based task processing methods described above during runtime.

[0020] In a fourth aspect, a storage medium is provided, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the large model-based task processing methods described above during runtime.

[0021] By means of the above technical scheme, the embodiment of the present application provides a task processing system and method based on a big model, a computer program product, and a storage medium, wherein the system includes a task initiator, a task simulator and a big model, wherein the task initiator is responsible for submitting a task request; the task simulator receives the task request, parses the task description information in the task request, identifies the key elements and logical structure of the task, and then constructs a directed acyclic graph based on the key elements and logical structure of the task, and generates a task set and an ordered task queue containing multiple subtasks according to the directed acyclic graph, and then calls the big model to obtain the processing results of each subtask in the task set, and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue; the big model, as an actual computing unit, receives the processed task request and generates a corresponding output, that is, Generate the processing results of each subtask and feed them back to the task simulator in real time; by decomposing a complex single task into a series of subtasks, each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model, which solves the problem of output token size limitation when the large model processes a complex single task, and there is no need to adjust the parameters of the large model or manual intervention, which improves the efficiency and accuracy of task processing; in addition, the quality and consistency of the output content are ensured through the task content and constraint information of each subtask, which meets the standards of academic or commercial documents; in addition, through the constructed directed acyclic graph, a task set containing multiple subtasks is generated, and through the directed acyclic graph, an ordered task queue that specifies the output order of the processing results of each subtask is generated, which ensures the sequentiality and logical consistency of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0023] Figure 1 An architecture diagram of a task processing system based on a large model provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of a directed acyclic graph provided by an embodiment of the present application is shown; Figure 3 A flowchart of a large model-based task processing method applied to a task simulator provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such use is interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to".

[0026] Before introducing the embodiments of the present application, the following technical terms are introduced first.

[0027] Large models refer to machine learning models with a large number of parameters and complex structures. They can process massive amounts of data and complete various complex tasks, such as natural language processing, computer vision, speech recognition, etc. Large models usually require a lot of computing resources and training data to train in order to better capture the complex relationships and patterns in the data.

[0028] A large language model is a type of large model that is specifically designed to handle natural language processing tasks. This type of model learns to understand the structure, semantics, context, and pragmatics of a language by training on a large text corpus. Large language models can handle a variety of natural language tasks, such as text classification, question-answering, and dialogue, and are an important tool for the path to artificial intelligence.

[0029] In order to solve the above technical problems, the embodiment of the present application provides a task processing system based on a large model, such as Figure 1 As shown, the big model-based task processing system 100 may include a task initiator 110 , a task simulator 120 and a big model 130 .

[0030] The task initiator 110 calls the interface of the task simulator 120 to submit a task request to the task simulator 120; The task simulator 120 receives the task request, parses the task description information in the task request, and identifies the key elements and logical structure of the task; The task simulator 120 constructs a directed acyclic graph based on the key elements and logical structure of the task, the directed acyclic graph including multiple subtask nodes with a logical order, the multiple subtask nodes are used to store the task content and constraint information of the multiple subtasks; generates a task set including multiple subtasks according to the directed acyclic graph; and generates an ordered task queue that specifies the output order of the processing results of each subtask according to the directed acyclic graph; wherein each subtask in the task set includes the task content and constraint information, and each subtask can be processed by the large model at one time, and the processing result of each subtask will not exceed the output limit of the large model; The task simulator 120 calls the interface of the large model 130 and sends the task content and constraint information of each subtask in the task set to the large model 130; The large model 130 generates processing results of each subtask according to the task content and constraint information of each subtask, and returns them to the task simulator 120; The task simulator 120 receives the processing results of each subtask, and returns the processing results of each subtask to the task initiator 110 according to the output order of the processing results of each subtask in the ordered task queue.

[0031] This embodiment decomposes a complex single task into a series of subtasks so that each subtask can be processed by the large model at one time. The processing results of each subtask will not exceed the output limit of the large model, which solves the problem of output token scale limitation when the large model processes a complex single task. There is no need to adjust the large model parameters or manual intervention, which improves the efficiency and accuracy of task processing. In addition, the quality and consistency of the output content are ensured through the task content and constraint information of each subtask, which meets the standards of academic or commercial documents. In addition, a task set containing multiple subtasks is generated through the constructed directed acyclic graph, and an ordered task queue that specifies the output order of the processing results of each subtask is generated through the directed acyclic graph, which ensures the sequentiality and logical consistency of task execution.

[0032] The present application provides a possible implementation method, and the constraint information may include format requirements, timeline historical background, content depth requirements, content breadth requirements, etc., which are not limited in the present embodiment. The present embodiment uses the task content and constraint information of each subtask to ensure that the large model can maintain consistency and coherence when generating the processing results of each subtask based on the task content and constraint information of each subtask.

[0033] The embodiment of the present application provides a possible implementation method. For a certain subtask in the task set, it may be called the current subtask; the constraint information may also include the summary information of other subtasks associated with the current subtask, and the summary information of other subtasks can be obtained through the task content of other subtasks. In this way, it can be ensured that each subtask is based on an appropriate context environment, thereby ensuring the consistency and coherence of the large model when generating the processing results of each subtask based on the task content and constraint information of each subtask.

[0034] A possible implementation method is provided in an embodiment of the present application, where a subtask node among multiple subtask nodes represents a subtask, the first subnode of the subtask node stores the task content of the subtask, and the second subnode of the subtask node stores the constraint information of the subtask.

[0035] In this embodiment, the number of the first sub-nodes may be one or more, and the number of the second sub-nodes may also be one or more, which is not limited in this embodiment.

[0036] Figure 2 A schematic diagram of a directed acyclic graph provided in an embodiment of the present application is shown. Figure 2 In the directed acyclic graph, a plurality of subtask nodes are included in a logical order, namely nodes 1, 2 to n, where n is a positive integer, node 1 represents subtask 1, node 2 represents subtask 2, and so on, and node n represents subtask n.

[0037] Continue to see Figure 2 The first child node A1 of node 1 represents the task content of subtask 1, and the second child node B1 of node 1 represents the constraint information of subtask 1; further, the child node B1-1 of the second child node B1 belongs to the private constraint information of subtask 1, and the child node B1-2 of the second child node B1 is connected to the public constraint information, which can be specifically connected to node C1.

[0038] The first child node A2 of node 2 represents the task content of subtask 2, and the second child node B2 of node 2 represents the constraint information of subtask 2; further, the child node B2-1 of the second child node B2 belongs to the private constraint information of subtask 2, and the child node B2-2 of the second child node B2 is connected to the public constraint information, which can be specifically connected to node C1.

[0039] By analogy, the first child node An of node n represents the task content of subtask n, and the second child node Bn of node n represents the constraint information of subtask n; further, the child node Bn-1 of the second child node Bn belongs to the private constraint information of subtask n, and the child node Bn-2 of the second child node Bn is connected to the public constraint information, which can be specifically connected to node C1.

[0040] As described above, constraint information may include format requirements, timeline historical background, content depth requirements, content breadth requirements, etc. The private constraint information and public constraint information here can be set according to actual needs, and this embodiment does not limit this. For example, format requirements can be used as public constraint information, and timeline historical background, content depth requirements, content breadth requirements, etc. can be used as private constraint information. It should be noted that the examples here are only illustrative and do not limit this embodiment.

[0041] In the embodiment of the present application, a possible implementation method is provided. The task simulator 120 calls the interface of the large model 130 to send the task content and constraint information of each subtask in the task set to the large model 130. Specifically, it can be: The task simulator 120 calls the interface of the large model 130, randomly selects the task content and constraint information of one or more subtasks from the task set according to the set period and sends them to the large model 130, until the task content and constraint information of all subtasks in the task set are sent to the large model 130. The set period here can be set according to actual needs, such as setting the period to 0.01 seconds or 0.001 seconds, etc., which is not limited in this embodiment.

[0042] This embodiment implements a dynamic, periodic task allocation and processing mechanism, so that the large model 130 can continuously receive and process each subtask from the task simulator 120, while maintaining the flexibility and configurability of the processing process.

[0043] In the embodiment of the present application, a possible implementation method is provided. If the output order of the processing results of each subtask in the ordered task queue is the processing result of the first subtask to the processing result of the Nth subtask, N is a positive integer; the task simulator 120 calls the interface of the large model 130 to send the task content and constraint information of each subtask in the task set to the large model 130, specifically, it can be: The task simulator 120 calls the interface of the large model 130, and based on the output order of the processing results of each subtask in the ordered task queue, sends the task content and constraint information of the first subtask in the task set, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask to the large model 130 in sequence.

[0044] This embodiment implements an orderly and continuous task processing flow, so that the large model 130 can process each subtask in the task set in sequence according to the output order of the processing results of each subtask in the ordered task queue, thereby improving the efficiency and accuracy of task processing.

[0045] The embodiment of the present application provides a possible implementation method, in which the task simulator 120 receives the processing results of each subtask and returns the processing results of each subtask to the task initiator 110 according to the output order of the processing results of each subtask in the ordered task queue, which may be: The task simulator 120 receives the processing results of each subtask, and according to the output order of the processing results of each subtask in the ordered task queue, if the processing result of the current subtask is received, and the processing result of the subtask before the current subtask has not been received, then it waits to receive the processing result of the subtask before the current subtask. When the processing result of the subtask before the current subtask is received, it returns the processing results of the subtask before the current subtask and the processing result of the current subtask to the task initiator 110 in sequence.

[0046] By waiting for and returning the processing results in sequence, the task simulator 120 in this embodiment improves the data consistency of the entire task processing flow, ensures that the task initiator 110 receives complete and sequentially arranged task results, and can provide a better user experience because the user can view the processing results in the logical order of the tasks instead of receiving an unordered result set, thereby improving the efficiency and reliability of the entire task processing flow.

[0047] The above introduces Figure 1 There are various implementation methods for each link of the illustrated embodiment. The task processing system based on a large model of the embodiment of the present application will be further explained below through specific embodiments.

[0048] In a specific embodiment, the user wants to generate an academic paper on "Future Urban Transportation". The following is the specific processing flow: 1) Task submission The task initiator 110 calls the interface of the task simulator 120 to submit a complex single task request to generate an academic paper on "Future Urban Transportation". The user provides a detailed background description, including the purpose of the research, the expected chapter structure (introduction, current situation analysis, future trend prediction, conclusion summary), and some key research questions and references.

[0049] 2) Task analysis After receiving the task request, the task simulator 120 firstly parses the task description information in the task request and extracts the following key elements and logical structures: Research purpose: To explore the future development direction of urban transportation and its impact on social economy; Expected chapter structure: introduction, current situation analysis, future trend forecast, conclusion summary; Key research questions: How to alleviate traffic congestion, improve public transportation efficiency, and reduce environmental pollution; References: Some authoritative literature and data sources provided by users.

[0050] 3) Task amplification and queue formation The task simulator 120 divides the paper writing task into several main stages and subdivides them into specific subtasks. Each subtask has clear task content and constraint information to ensure that they can maintain consistency and coherence during execution. The task simulator 120 constructs a directed acyclic graph based on the key elements and logical structure of the task, and then the subtasks in the task set generated according to the directed acyclic graph are as follows: (1) Introduction writing (first level directory) Task content: Introduce the research background, purpose and significance, and outline the structure of the paper.

[0051] Constraint information: 1,000 words, an overview of each section of the first-level directory.

[0052] (2) Current situation analysis (first level directory) (a) Public transport development (secondary directory) Task content: Analyze the current status of urban public transportation systems, including subways, buses, light rail, etc.

[0053] Constraint information: format requirements, overview of each part of the first-level directory, and overview of other parts of the second-level directory where this task is located.

[0054] (b) Private car usage (secondary directory) Task content: Explore the popularity and usage habits of private cars and their impact on the transportation system.

[0055] Constraint information: format requirements, overview of each part of the first-level directory, and overview of other parts of the second-level directory where this task is located.

[0056] (c) Traffic congestion problem (secondary directory) Task content: Analyze the main causes and impacts of current urban traffic congestion.

[0057] Constraint information: format requirements, overview of each part of the first-level directory, and overview of other parts of the second-level directory where this task is located.

[0058] (3) Future trend forecast (first level directory) (a) Intelligent Transportation Systems (Secondary Category) Task content: Discuss the development of intelligent transportation technologies, such as autonomous driving and vehicle networking.

[0059] Constraint information: format requirements, overview of each part of the first-level directory, and overview of other parts of the second-level directory where this task is located.

[0060] (b) Sustainable transport modes (secondary category) Task content: Explore future sustainable transportation modes, such as shared travel, electric vehicles, etc.

[0061] Constraint information: format requirements, overview of each part of the first-level directory, and overview of other parts of the second-level directory where this task is located.

[0062] (4) Conclusion and summary (first level directory) Task content: Summarize the full text and put forward suggestions and prospects for the future development of urban transportation.

[0063] Constraint information: format requirements, overview of each part of the first-level directory.

[0064] 4) Task queuing The task simulator 120 generates an ordered task queue that specifies the output order of the processing results of each subtask according to the directed acyclic graph. The specific order is as follows: ①Writing the introduction; ② Development of public transportation; ③Private car usage; ④ Traffic congestion; ⑤ Intelligent transportation system; ⑥ Sustainable transportation mode; ⑦Conclusion and summary.

[0065] 5) Task Execution The task simulator 120 calls the interface of the large model 130 and sends the task content and constraint information of each subtask in the task set to the large model 130. Here, the task simulator 120 calls the interface of the large model 130 and can randomly select the task content and constraint information of one or more subtasks from the task set according to a set period and send them to the large model 130 until the task content and constraint information of all subtasks in the task set are sent to the large model 130.

[0066] Alternatively, the task simulator 120 may call the interface of the large model 130, and based on the output order of the processing results of each subtask in the ordered task queue, send the task content and constraint information of the first subtask in the task set, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask in sequence to the large model 130.

[0067] After receiving the task content and constraint information of each subtask, the large model 130 immediately starts processing and generates corresponding output based on the provided context.

[0068] The following is the specific implementation process: ① Introduction writing: The task simulator 120 sends the subtask of writing the introduction to the large model 130 , along with the background description and research purpose provided by the user. The large model 130 generates the introduction part and returns the result to the task simulator 120 .

[0069] ② Public transportation development: The task simulator 120 sends the subtask of public transportation development to the big model 130. The big model 130 generates the content of public transportation development and returns the result to the task simulator 120.

[0070] ③Private car usage: The task simulator 120 sends the subtask of the private car usage to the large model 130. The large model 130 generates the content of the private car usage and returns the result to the task simulator 120.

[0071] ④ Traffic congestion problem: The task simulator 120 sends the subtask of the traffic congestion problem to the large model 130. The large model 130 generates the content of the traffic congestion problem and returns the result to the task simulator 120.

[0072] ⑤ Intelligent Transportation System: The task simulator 120 sends the subtasks of the intelligent transportation system to the large model 130. The large model 130 generates the contents of the intelligent transportation system and returns the results to the task simulator 120.

[0073] ⑥ Sustainable transportation mode: The task simulator 120 sends the subtask of the sustainable transportation mode to the big model 130. The big model 130 generates the content of the sustainable transportation mode and returns the result to the task simulator 120.

[0074] ⑦ Conclusions and Summary: The task simulator 120 sends the subtask of the conclusion summary to the large model 130. The large model 130 generates the content of the conclusion summary and returns the result to the task simulator 120.

[0075] 6) Results returned The task simulator 120 receives the processing results of each subtask, and returns the processing results of each subtask to the task initiator 110 according to the output order of the processing results of each subtask in the ordered task queue.

[0076] The task simulator 120 introduced in this embodiment can effectively expand the output capability of the large model 130 without making any changes to the large model 130 itself. By decomposing complex tasks into a series of simple subtasks, the task simulator 120 ensures the quality and consistency of the output content. In addition, the task simulator 120 is also highly flexible and adaptable, and can adjust its working mode according to different task types and user needs, thereby improving the versatility and applicability of the system.

[0077] Based on the large model-based task processing system provided in the above embodiments and based on the same inventive concept, an embodiment of the present application also provides a large model-based task processing method applied to a task simulator.

[0078] Figure 3 The flowchart of the task processing method based on a large model applied to a task simulator provided in an embodiment of the present application is shown. Figure 3 As shown, the large model-based task processing method applied to the task simulator may include the following steps S301 to S305: Step S301, receiving a task request submitted by a task initiator by calling an interface of a task simulator; Step S302, parsing the task description information in the task request to identify the key elements and logical structure of the task; Step S303, based on the key elements and logical structure of the task, construct a directed acyclic graph, the directed acyclic graph includes multiple subtask nodes with a logical order, and the multiple subtask nodes are used to store the task content and constraint information of the multiple subtasks; according to the directed acyclic graph, generate a task set including multiple subtasks; and according to the directed acyclic graph, generate an ordered task queue that specifies the output order of the processing results of each subtask; wherein each subtask in the task set includes task content and constraint information, and each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model; Step S304, calling the interface of the big model, and sending the task content and constraint information of each subtask in the task set to the big model; Step S305, receiving the processing results of each subtask returned by the large model, and returning the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue.

[0079] This embodiment decomposes a complex single task into a series of subtasks so that each subtask can be processed by the large model at one time. The processing results of each subtask will not exceed the output limit of the large model, which solves the problem of output token scale limitation when the large model processes a complex single task. There is no need to adjust the large model parameters or manual intervention, which improves the efficiency and accuracy of task processing. In addition, the quality and consistency of the output content are ensured through the task content and constraint information of each subtask, which meets the standards of academic or commercial documents. In addition, a task set containing multiple subtasks is generated through the constructed directed acyclic graph, and an ordered task queue that specifies the output order of the processing results of each subtask is generated through the directed acyclic graph, which ensures the sequentiality and logical consistency of task execution.

[0080] The present application provides a possible implementation method, and the constraint information may include format requirements, timeline historical background, content depth requirements, content breadth requirements, etc., which are not limited in the present embodiment. The present embodiment uses the task content and constraint information of each subtask to ensure that the large model can maintain consistency and coherence when generating the processing results of each subtask based on the task content and constraint information of each subtask.

[0081] The embodiment of the present application provides a possible implementation method. For a certain subtask in the task set, it may be called the current subtask; the constraint information may also include the summary information of other subtasks associated with the current subtask, and the summary information of other subtasks can be obtained through the task content of other subtasks. In this way, it can be ensured that each subtask is based on an appropriate context environment, thereby ensuring the consistency and coherence of the large model when generating the processing results of each subtask based on the task content and constraint information of each subtask.

[0082] A possible implementation method is provided in an embodiment of the present application, where a subtask node among multiple subtask nodes represents a subtask, the first subnode of the subtask node stores the task content of the subtask, and the second subnode of the subtask node stores the constraint information of the subtask.

[0083] In this embodiment, the number of the first sub-nodes may be one or more, and the number of the second sub-nodes may also be one or more, which is not limited in this embodiment.

[0084] The embodiment of the present application provides a possible implementation method. The above step S304 calls the interface of the big model to send the task content and constraint information of each subtask in the task set to the big model. Specifically, the following steps S304-1 may be included: Step S304-1, calling the interface of the large model, randomly selecting the task content and constraint information of one or more subtasks from the task set according to the set period and sending them to the large model, until the task content and constraint information of all subtasks in the task set are sent to the large model. The set period here can be set according to actual needs, such as setting the period to 0.01 seconds or 0.001 seconds, etc., which is not limited in this embodiment.

[0085] This embodiment implements a dynamic, periodic task allocation and processing mechanism, so that the large model can continuously receive and process various subtasks from the task simulator, while maintaining the flexibility and configurability of the processing process.

[0086] The embodiment of the present application provides a possible implementation method. If the output order of the processing results of each subtask in the ordered task queue is the processing result of the first subtask to the processing result of the Nth subtask, N is a positive integer; the above step S304 calls the interface of the large model to send the task content and constraint information of each subtask in the task set to the large model, which may specifically include the following steps S304-2: Step S304-2, call the interface of the big model, and based on the output order of the processing results of each subtask in the ordered task queue, send the task content and constraint information of the first subtask in the task set, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask to the big model in sequence.

[0087] This embodiment implements an orderly and continuous task processing flow, so that the large model can process each subtask in the task set in sequence according to the output order of the processing results of each subtask in the ordered task queue, thereby improving the efficiency and accuracy of task processing.

[0088] The embodiment of the present application provides a possible implementation method. The above step S305 receives the processing results of each subtask returned by the large model, and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue. Specifically, the following steps S305-1 may be included: Step S305-1, receive the processing results of each subtask returned by the large model, and according to the output order of the processing results of each subtask in the ordered task queue, if the processing result of the current subtask is received, and the processing result of the subtask before the current subtask has not been received, then wait to receive the processing result of the subtask before the current subtask. When the processing result of the subtask before the current subtask is received, return the processing results of the subtask before the current subtask and the processing result of the current subtask to the task initiator in sequence.

[0089] By waiting for and returning the processing results in sequence, the task simulator in this embodiment improves the data consistency of the entire task processing flow, ensures that the task initiator receives complete and sequentially arranged task results, and can provide a better user experience because the user can view the processing results in the logical order of the tasks instead of receiving an unordered result set, thereby improving the efficiency and reliability of the entire task processing flow.

[0090] Based on the same inventive concept, an embodiment of the present application further provides a computer program product, including a computer program, wherein the computer program is configured to execute the large model-based task processing method of any of the above embodiments when running.

[0091] Based on the same inventive concept, an embodiment of the present application further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute any one of the large model-based task processing methods of the above-mentioned embodiments when running.

[0092] Those skilled in the art can clearly understand that the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the aforementioned method embodiments, and for the sake of brevity, they are not further described here.

[0093] Those skilled in the art can understand that the technical solution of the present application can be essentially or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, which includes a number of program instructions, so that an electronic device (such as a personal computer, a server, or a network device, etc.) executes all or part of the steps of the method described in each embodiment of the present application when running the program instructions. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0094] Alternatively, all or part of the steps of implementing the aforementioned method embodiments may be accomplished by hardware associated with program instructions (such as electronic devices such as personal computers, servers, or network devices), and the program instructions may be stored in a computer-readable storage medium. When the program instructions are executed by a processor of an electronic device, the electronic device executes all or part of the steps of the methods described in the embodiments of the present application.

[0095] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that within the spirit and principles of the present application, they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate from the protection scope of the present application.

Claims

1. A task processing system based on a large model, characterized in that: The system includes a task initiator, a task simulator and a large model; The task initiator calls the interface of the task simulator to submit a task request to the task simulator; The task simulator receives the task request, parses the task description information in the task request, and identifies the key elements and logical structure of the task; The task simulator constructs a directed acyclic graph based on the key elements and logical structure of the task, wherein the directed acyclic graph includes a plurality of subtask nodes having a logical order, and the plurality of subtask nodes are used to store the task contents and constraint information of the plurality of subtasks; According to the directed acyclic graph, a task set including a plurality of subtasks is generated; and according to the directed acyclic graph, an ordered task queue is generated that specifies the output order of the processing results of each subtask; wherein each subtask in the task set includes task content and constraint information, and each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model; The task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model; The big model generates the processing results of each subtask according to the task content and constraint information of each subtask, and returns them to the task simulator; The task simulator receives the processing results of each subtask, and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue.

2. The system according to claim 1, characterized in that The constraint information includes one or more of format requirements, timeline historical background, content depth requirements, and content breadth requirements.

3. The system according to claim 2, characterized in that For a subtask in the task set, we may call it the current subtask; The constraint information also includes summary information of other subtasks associated with the current subtask, and the summary information of other subtasks is obtained through the task contents of other subtasks.

4. The system according to claim 1, characterized in that A subtask node among the multiple subtask nodes represents a subtask, a first subnode of the subtask node stores the task content of the subtask, and a second subnode of the subtask node stores the constraint information of the subtask.

5. The system according to claim 1, characterized in that The task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model, including: The task simulator calls the interface of the large model, randomly selects the task content and constraint information of one or more subtasks from the task set according to the set period and sends them to the large model until the task content and constraint information of all subtasks in the task set are sent to the large model.

6. The system according to claim 1, characterized in that If the output order of the processing results of each subtask in the ordered task queue is the processing result of the first subtask to the processing result of the Nth subtask, N is a positive integer; The task simulator calls the interface of the large model and sends the task content and constraint information of each subtask in the task set to the large model, including: The task simulator calls the interface of the large model, and based on the output order of the processing results of each subtask in the ordered task queue, sends the task content and constraint information of the first subtask in the task set, the task content and constraint information of the second subtask, and the task content and constraint information of the Nth subtask to the large model in sequence.

7. The system according to claim 5 or 6, characterized in that: The task simulator receives the processing results of each subtask and returns the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue, including: The task simulator receives the processing results of each subtask, and according to the output order of the processing results of each subtask in the ordered task queue, if the processing result of the current subtask is received, and the processing result of the subtask before the current subtask has not been received, it waits to receive the processing result of the subtask before the current subtask. When the processing result of the subtask before the current subtask is received, the processing results of the subtask before the current subtask and the processing result of the current subtask are returned to the task initiator in sequence.

8. A task processing method based on a large model, characterized in that: Applied to a mission simulator, the method comprises: Receive a task request submitted by a task initiator by calling an interface of a task simulator; Parsing the task description information in the task request to identify the key elements and logical structure of the task; Based on the key elements and logical structure of the task, a directed acyclic graph is constructed, wherein the directed acyclic graph includes a plurality of subtask nodes with a logical order, and the plurality of subtask nodes are used to store the task content and constraint information of the plurality of subtasks; according to the directed acyclic graph, a task set including the plurality of subtasks is generated; and according to the directed acyclic graph, an ordered task queue is generated that specifies the output order of the processing results of each subtask; wherein each subtask in the task set includes the task content and constraint information, and each subtask can be processed by the large model at one time, and the processing results of each subtask will not exceed the output limit of the large model; Call the interface of the big model to send the task content and constraint information of each subtask in the task set to the big model; Receive the processing results of each subtask returned by the large model, and return the processing results of each subtask to the task initiator according to the output order of the processing results of each subtask in the ordered task queue.

9. A computer program product, comprising a computer program, characterized in that The computer program is configured to execute the large model-based task processing method according to claim 8 when running.

10. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the large model-based task processing method according to claim 8 when running.

Images