Information processing methods, devices, electronic equipment, and storage media based on large models

Abstract

This disclosure provides a method, apparatus, electronic device, and storage medium for information processing based on a large model, relating to the field of artificial intelligence technology, particularly to the fields of large models, intelligent question answering, intelligent search, and smart healthcare. The method for information processing based on a large model includes: orchestrating tasks based on demand information related to a target object to obtain a task orchestration strategy containing multiple initial tasks; updating at least one initial task in the task orchestration strategy based on the target object's interactive operations during the execution of the task orchestration strategy using the large model to obtain a target task; and determining feedback information matching the target object's demand intent based on the target execution result determined by executing the target task using the large model.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to the fields of large models, intelligent question answering, intelligent search, and smart healthcare. Background Technology

[0002] Large Language Models (LLMs) are models built using deep learning algorithms that can perform generative tasks based on natural language understanding. LLMs can be used to process user input questions and meet user needs by generating responses based on natural language expressions. Summary of the Invention

[0003] This disclosure provides an information processing method, apparatus, electronic device, and storage medium based on a large model.

[0004] According to one aspect of this disclosure, a large-scale model-based information processing method is provided, comprising: performing task orchestration on requirement information related to a target object to obtain a task orchestration strategy containing multiple initial tasks; updating at least one initial task in the task orchestration strategy based on the interactive operations of the target object during the execution of the task orchestration strategy using the large-scale model to obtain a target task; and determining feedback information matching the requirement intent of the target object based on the target execution result determined by executing the target task using the large-scale model.

[0005] According to another aspect of this disclosure, a large-scale model-based information processing apparatus is provided, comprising: a first obtaining module, configured to perform task orchestration on demand information related to a target object to obtain a task orchestration strategy containing multiple initial tasks; a second obtaining module, configured to update at least one initial task in the task orchestration strategy based on the interactive operations of the target object during the execution of the task orchestration strategy using the large-scale model to obtain a target task; and a feedback information determining module, configured to determine feedback information matching the demand intention of the target object based on the target execution result determined by executing the target task using the large-scale model.

[0006] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method provided according to an embodiment of this disclosure.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform a method provided according to an embodiment of this disclosure.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided according to embodiments of this disclosure.

[0009] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0010] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0011] Figure 1 This illustration schematically shows an exemplary system architecture for applying large-model-based information processing methods and apparatus according to embodiments of the present disclosure;

[0012] Figure 2 A flowchart illustrating an information processing method based on a large model according to an embodiment of the present disclosure is shown schematically.

[0013] Figure 3 The illustration schematically shows a principle diagram of a task orchestration strategy according to an embodiment of the present disclosure;

[0014] Figure 4 The schematic diagram illustrates the principle of a large-model-based information processing method according to an embodiment of the present disclosure;

[0015] Figure 5 This illustration schematically shows a system architecture diagram of an information processing system according to an embodiment of the present disclosure;

[0016] Figure 6 A flowchart illustrating a large-model-based information processing method according to another embodiment of the present disclosure is shown schematically.

[0017] Figure 7 A block diagram of a large-model-based information processing apparatus according to an embodiment of the present disclosure is shown schematically.

[0018] Figure 8 A schematic diagram illustrating the structure of an intelligent agent of artificial intelligence according to embodiments of the present disclosure; and

[0019] Figure 9 A schematic block diagram of an example electronic device is shown that can be used to implement the large-model-based information processing method of embodiments of the present disclosure. Detailed Implementation

[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] In the technical solution disclosed herein, the acquisition, storage, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0022] The inventors discovered that during question-and-answer interactions with users based on large models, the large models take a long time to execute text generation tasks, and the generated responses often differ significantly from the user's actual needs and intentions. Users are required to modify their questions multiple times based on the responses or click "regenerate" to instruct the large model to generate new responses. This results in long user wait times during intelligent question-and-answer interactions with large models and can also lead to significant redundant computational overhead on the servers and other computing devices deploying the large models.

[0023] Embodiments of this disclosure provide a large-model-based information processing method, apparatus, electronic device, and storage medium. The large-model-based information processing method includes: orchestrating tasks based on requirement information related to a target object to obtain a task orchestration strategy comprising multiple initial tasks; updating at least one initial task in the task orchestration strategy based on the target object's interactive operations during the execution of the task orchestration strategy using the large model to obtain a target task; and determining feedback information matching the target object's requirement intent based on the target execution result determined by executing the target task using the large model.

[0024] According to embodiments of this disclosure, by orchestrating tasks based on the target object's demand information, the execution results of multiple initial tasks in the task orchestration strategy can satisfy the demand attributes represented by the demand information. During the execution of multiple initial tasks in the task orchestration strategy using a large model, the initial tasks in the task orchestration strategy are synchronously updated based on the target object's interactive operations. This allows for dynamic adjustment of the initial tasks in the task orchestration strategy according to the real-time demands expressed by the user's interactive operations during the execution of initial tasks based on the large model. This avoids the waiting time for the user to wait for all initial tasks in the task orchestration strategy to be completed before updating the demand information to obtain a response that meets the user's actual needs. Simultaneously, the target execution results updated based on the intent of the interactive operations can satisfy the target object's real-time demands. Furthermore, by adjusting the task orchestration logic generated for the target object's actual demands in real time, a target execution result matching the target object's actual demands can be determined. Thus, the feedback information determined based on the target execution results can satisfy the target object's actual demands even before the task orchestration logic is completed, improving the efficiency and quality of feedback information generation, reducing redundant computational overhead on computing devices used to deploy large models, and enhancing user satisfaction and user experience.

[0025] Figure 1 The illustration schematically shows an exemplary system architecture for applying large-model-based information processing methods and apparatus according to embodiments of the present disclosure.

[0026] It is important to note that Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture for applying the information processing method and apparatus based on large models may include a terminal device, but the terminal device can implement the information processing method and apparatus based on large models provided by the embodiments of this disclosure without interacting with the server.

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

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

[0029] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0030] Server 105 can be a server that provides various services, such as a backend management server that supports the content browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0031] It should be noted that the information processing method based on a large model provided in this disclosure can generally be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103. Correspondingly, the information processing device based on a large model provided in this disclosure can also be disposed in the first terminal device 101, the second terminal device 102, or the third terminal device 103.

[0032] Alternatively, the large-model-based information processing method provided in this embodiment can generally be executed by server 105. Correspondingly, the large-model-based information processing apparatus provided in this embodiment can generally be located in server 105. The large-model-based information processing method provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the large-model-based information processing apparatus provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0033] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0034] Figure 2 A flowchart illustrating a large-model-based information processing method according to an embodiment of the present disclosure is shown schematically.

[0035] like Figure 2 As shown, the information processing method based on the large model includes operations S210~S230.

[0036] In operation S210, task orchestration is performed on the requirement information related to the target object to obtain a task orchestration strategy containing multiple initial tasks.

[0037] During operation S220, in the process of executing the task orchestration strategy using a large model, at least one initial task in the task orchestration strategy is updated based on the interaction operation of the target object to obtain the target task.

[0038] In operation S230, based on the target execution results determined by using the large model to execute the target task, feedback information that matches the needs and intentions of the target object is determined.

[0039] According to embodiments of this disclosure, the requirement information can be information characterizing the target object's requirements, such as question text or query information input by the target object. The requirement information can include information of any data type, such as text, images, and audio; embodiments of this disclosure do not limit the specific data type of the requirement information.

[0040] According to embodiments of this disclosure, task orchestration of demand information may include processing the demand information based on a deep learning model to generate a task orchestration strategy that includes multiple initial tasks and dependencies between the initial tasks. Initial tasks may include any type of data processing task such as information search, image recognition, or data computation. Dependencies may represent the execution logic between different initial tasks.

[0041] According to embodiments of this disclosure, a large model can be used to execute multiple initial tasks based on dependencies in a task orchestration strategy. For example, the large model can execute a first initial task, "image segmentation task," to segment the image in the requirement information, obtaining a segmented image region of building A. The image region of building A can be the initial execution result of the first initial task, "image segmentation task." After obtaining the image region of building A, a second initial task, "object recognition task," which has a dependency on the first initial task, can be determined based on dependencies in the task orchestration strategy. By using the large model to call an object recognition tool to execute the second initial task, the object type obtained by recognizing the image region of building A can be determined to be "Gymnasium A."

[0042] In some embodiments, the task orchestration strategy can be represented based on a topology such as a task chain or task tree containing multiple initial tasks. The topology representing the task orchestration strategy can include task nodes representing initial tasks and edge relationships representing dependencies. This allows the large model to execute multiple initial tasks more clearly according to the execution logic indicated by the dependencies in the task orchestration strategy, thereby realizing the execution of task orchestration logic using the large model.

[0043] According to embodiments of this disclosure, the large model can be a trained large language model, which can process multimodal data such as text, speech, and images based on a large number of model parameters to perform the initial task. It should be noted that the number of large models involved in the embodiments of this disclosure can be one or more, and the embodiments of this disclosure do not limit the number of large models.

[0044] According to embodiments of this disclosure, executing task orchestration logic using a large model may include using a computing unit that has deployed the model parameters to execute multiple initial tasks according to the dependencies indicated by the task orchestration logic. The process of executing task orchestration logic using a large model may involve the computing unit performing data processing on the initial tasks.

[0045] In some embodiments, during the execution of task orchestration logic using a large model, feedback information matching the needs of the target object has not yet been output. The target object can perform any type of interactive operation, such as input operations, during the execution of the initial task using the large model. The operation type of the interactive operation and the input information carried by the interactive operation can characterize the real-time demand attributes of the target object during the execution of the task orchestration logic.

[0046] In some embodiments, updating at least one initial task in the task orchestration logic based on the interactive operation of the target object may include updating the task attribute parameters of the initial task based on the operation information carried by the interactive operation of the target object, thereby obtaining the target task. For example, tool attribute parameters such as tool name and tool parameters of the tool resources involved in executing the initial task may be updated, but it is not limited to this. The task attribute parameters may also include parameters used to control the task execution result, such as the reference knowledge domain, style attributes, and data format of the output result related to the initial task.

[0047] For example, updating the initial task could include updating the prompt for performing the initial task "information retrieval" using the large model based on the interactive input "please refer to medical domain knowledge", and performing the target search task based on the new prompt "search for medical domain knowledge".

[0048] In some embodiments, updating at least one initial task in the task orchestration logic based on the interactive operation of the target object may further include updating the dependencies between multiple initial tasks in the task orchestration strategy. For example, the serial execution logic of initial tasks 1 to 6 in the task orchestration strategy can be modified to execute the first task chain of initial tasks 1 to 4 and the second task chain of initial tasks 5 to 6 in parallel, based on the operation information indicated by the interactive operation. This allows the execution logic and execution order between multiple initial tasks to be adjusted by modifying the dependencies in the task orchestration logic based on the interactive operation, enabling the large model to execute multiple tasks in multiple task chains based on a parallel model, thereby improving the efficiency of generating feedback information. It should be understood that the initial task after updating the dependencies can be understood as the target task. Feedback information can be determined based on the target execution result of the target task and the initial execution result of the initial task.

[0049] In some embodiments, updating at least one initial task in the task orchestration logic based on the interactive operation of the target object may further include updating the task attribute parameters of the initial tasks in the task orchestration logic and the dependencies between the initial tasks. The embodiments of this disclosure will not be described in detail here.

[0050] According to embodiments of this disclosure, by updating at least one initial task in the task orchestration logic based on the interactive operations of the target object, the updated target task can meet the actual needs of the interactive operation instructions. Thus, by utilizing a large model to execute the target task in the task orchestration logic and obtaining the target execution result, the actual needs of the target object can be met, ensuring that the feedback information determined based on the target execution result matches the real-time needs of the target object. Simultaneously, it avoids the need for the target object to input modified requirement information and then reuse the large model to respond to the target object after the task orchestration logic has fully executed, reducing the waiting time for the target object to obtain satisfactory feedback, improving user experience, and reducing the computational overhead of the computing devices deploying the large model.

[0051] In some embodiments, the information processing method based on a large model may further include: pushing at least one of a task orchestration strategy and the initial execution result of an initial task to the target object.

[0052] In one example, a task orchestration strategy can be pushed to a target object, and a directed acyclic graph (DAG) representing multiple initial tasks and their dependencies can be displayed. This allows the target object to perform interactive operations on the task nodes representing the initial tasks and the edge relationships representing the dependencies in the DAG to input operation information.

[0053] In one example, the execution results of the initial task in the task orchestration strategy, which utilizes a large model, can be pushed to the target object. For instance, multiple landscape images of city A retrieved by the initial task "image retrieval task" can be pushed to the target object. The target object can remove specified landscape images based on the displayed landscape images to reduce the amount of data processing generated by subsequent initial or target tasks in the task orchestration strategy, reduce the computational overhead of generating feedback information, and minimize the mismatch between the generated feedback information and the target object's actual needs due to processing landscape images that are not needed by the target object, thereby improving the user experience.

[0054] In some embodiments, initial execution results and task orchestration strategies can also be pushed to the target object so that the target object can flexibly adjust and orchestrate the tasks that the large model needs to execute, thereby enabling the task orchestration strategy to more accurately meet the needs of the target object, improve the matching degree between feedback information and the target object's needs and intentions, and enhance the user experience.

[0055] According to embodiments of this disclosure, the first obtaining module may include: using a large model to perform semantic understanding of the demand information to obtain the demand intent; and performing task orchestration based on the demand intent to obtain a task orchestration strategy.

[0056] According to embodiments of this disclosure, the demand intent can represent the demand attributes of the target object. For example, the demand intent can be unstructured natural language description information such as "generate a press release with the theme of a campus sports meet." Alternatively, the demand intent can be structured data, such as multiple intent keywords represented in a table. Utilizing a large model to understand the demand information can correct semantic ambiguities or fill in missing information, thereby ensuring that the obtained demand intent accurately represents the actual demand attributes of the target object. This improves the accuracy of subsequent task scheduling strategies, reduces the decrease in the matching degree between feedback information and the target object's needs due to missing tasks, and ultimately enhances the user experience.

[0057] In some embodiments, using a large model to perform semantic understanding of demand information to obtain demand intent may further include: determining the initial task corresponding to each of the multiple intent contents in the structured demand intent, and then determining the task orchestration strategy based on a preset dependency relationship.

[0058] In some embodiments, using a large model to perform semantic understanding of demand information to obtain demand intent may also include: using a large model to process demand information and target object attribute information, and then using the large model to integrate the target object attributes such as preferences and habits represented by demand information and object attribute information to perform demand intent understanding, thereby obtaining demand intent that can meet the actual needs of the target object and improve user experience.

[0059] In some embodiments, using a large model to perform semantic understanding of demand information to obtain demand intent may also include: using a large model based on object attribute knowledge graphs to perform semantic understanding of demand information to obtain demand intent.

[0060] According to embodiments of this disclosure, object attribute nodes in the object attribute knowledge graph represent the historical interaction information or object preference attributes of the target object. Historical interaction information can represent, but is not limited to, the target object's browsing, collection, liking, and search interactions during historical periods. Historical interaction information can also include contextual content input by the target object, historical dialogue content with the large model, etc. Object preference attributes can include attribute information determined based on the target object's preference selection operations, such as sports preference attributes, medical knowledge preference attributes, etc. Embodiments of this disclosure do not limit the specific type or generation method of object preference attributes.

[0061] According to embodiments of this disclosure, an object indicator graph is constructed by analyzing historical interaction information and object preference attributes. This allows for the representation of a target object's personalized preferences or habits based on entities, and the representation of relationships between various types of personalized preferences or habits based on edges. This enables the structured representation of a target object's personalized potential needs based on an object attribute knowledge graph. A large model is then used to semantically understand the demand information through the object knowledge graph. This allows the large model to clearly understand the target object's potential demand attributes based on the structured representation of personalized preferences. Furthermore, by combining personalized preference attributes with the understanding of the demand intent contained in the demand information, the large model can more accurately represent the target object's actual intent, thereby improving the rationality and adaptability of subsequent task orchestration strategies. Finally, the updated task orchestration logic enhances the user experience.

[0062] In some embodiments, the demand intent includes a desired time intent that characterizes the target object's expectation of receiving feedback information. The desired time intent can represent an intent related to the time of receiving feedback information, such as the expected moment, duration, or time period. For example, the desired time intent could be "20 minutes later" or "11:00 AM on [date]". The demand intent containing the desired time intent can be obtained by utilizing the first major model to perform intent understanding on the demand information.

[0063] In some embodiments, task orchestration based on demand intent to obtain a task orchestration strategy may include: using the delay prompt information determined by the expected time intent, and employing the first major model to orchestrate the demand content of the demand intent to obtain a task orchestration strategy.

[0064] According to embodiments of this disclosure, latency warning information can be used to indicate the strategy execution time required for the large model to execute task orchestration logic to output feedback information, and to meet the time or duration at which the target object expects to obtain feedback information.

[0065] For example, a delay warning message could be "Output a PPT file related to this month's work report within 20 minutes".

[0066] In some embodiments, a first large model orchestrates tasks according to the intended requirements, while a second large model executes the task orchestration strategy. The first and second large models can be different large models with different model parameters. By collaborating based on multiple different large models to perform task orchestration and execute task orchestration strategies separately, different types of tasks can be executed based on the model capabilities of each large model. This collaboration of multiple large models improves the efficiency and quality of task orchestration strategy generation and execution, thereby enhancing the matching degree between feedback information and target objects and improving the user experience.

[0067] In some embodiments, the execution time of the second major model in executing the task orchestration strategy to obtain feedback information is matched with the expected time intent. For example, the time required for the second major model to obtain feedback information by executing the task orchestration strategy of the first major model based on the time delay prompt information is matched with the time requirement attributes such as the expected time, duration, or time period of obtaining feedback information as represented by the expected time intent. Thus, the task orchestration strategy of the first major model that adapts to the second major model to output feedback information according to the expected time intent can meet the time requirements of the target object to obtain feedback information, thereby satisfying the diverse actual needs of users and improving the user experience.

[0068] In some embodiments, the large model-based information processing method may further include: based on the task attribute information of the initial task, using the first large model to perform semantic understanding according to the model description information of multiple candidate large models, and determining the second large model for performing the initial task from the multiple candidate large models.

[0069] According to embodiments of this disclosure, the model description information characterizes the model performance attributes of the candidate large model. For example, model performance attributes can be model performance metrics such as accuracy, average latency, and computational cost. Furthermore, model performance attributes can also include the domain to which the candidate large model is adapted, such as indicating that the candidate large model is adapted to various types of domains such as medical, educational, and engineering construction. As another example, model performance attributes can also represent the scenarios to which the candidate large model is adapted, such as indicating that the candidate large model is adapted to intelligent search scenarios, script editing scenarios, and marketing copywriting scenarios.

[0070] In some embodiments, model performance attributes may also represent model performance metrics used to evaluate the computational cost of candidate large models, such as the scale of model parameters.

[0071] According to embodiments of this disclosure, by semantically understanding the model description information of multiple candidate large models based on the task attribute information of multiple initial tasks, the first large model can fully understand the degree of matching between the model performance of the candidate large models and the performance requirements for executing the initial tasks. This allows the second model determined from the candidate large models for each initial task to more accurately adapt to the task attribute information such as the task requirement scenario and task execution parameters of the initial tasks in the task orchestration strategy. This further improves the ability of the execution results output by the task orchestration strategy to match the intent represented by the requirement information, thereby reducing the frequency of interactive operations performed by the target object, increasing the degree of matching between the feedback information and the target object's intent, and thus improving the user experience.

[0072] In one embodiment, multiple initial tasks in the orchestration strategy can be executed based on different second-large models. Thus, the execution efficiency of the initial tasks can be improved by using the task orchestration strategy output by the first-large model to schedule multiple different large models to execute different initial tasks.

[0073] In some embodiments, the orchestration strategy includes multiple task topologies that are dependent on a specified initial task, and the interaction operation is associated with a first task topology among the multiple task topologies.

[0074] According to embodiments of this disclosure, a task topology can be represented as any topology such as a task chain or a task tree. Multiple initial tasks within a task topology have dependencies, and the task execution process can be represented by the dependencies between different task topologies and specified initial tasks.

[0075] Figure 3 The illustration shows a schematic diagram of the principle of a task orchestration strategy according to an embodiment of the present disclosure.

[0076] like Figure 3As shown, the task orchestration strategy in this embodiment can be a file task orchestration strategy 300 for generating presentation files. By utilizing the requirement information "generate a presentation file introducing the attractions of city A" input by the first large model, task orchestration is performed to obtain the file task orchestration strategy 300. The file task orchestration strategy 300 can include multiple initial tasks, which can be based on... Figure 3 The circular elements shown represent the initial tasks. These are designated as Initial Task 1 through Initial Task 6. Initial Task 1 represents the presentation outline generation task, Initial Task 2 represents the image material retrieval task, Initial Task 3 represents the image material editing task, Initial Task 4 represents the scenic spot description text retrieval task, Initial Task 5 represents the scenic spot description text semantic fusion task, and Initial Task 6 represents the presentation file creation task.

[0077] The file task orchestration strategy 300 may include two different task topologies that are dependent on the first initial task, namely the first task topology 310 and the second task topology 320. Both the first task topology 310 and the second task topology 320 are dependent on the first initial task.

[0078] In some embodiments, updating at least one initial task in the task orchestration strategy based on the interaction of the target object may include: using a large model to understand the intent of the interaction information and updating the first initial task related to the operation information in the first task topology.

[0079] For example, the first initial task is a page retrieval task, and the operation information can carry the input information of "Website A". The task execution parameters of the page retrieval task can be modified based on "Website A" so that by executing the target task, a page retrieval for "Website A" is added, and the target execution result includes pages from Website A related to the user's intent.

[0080] According to embodiments of this disclosure, feedback information is determined based on the initial execution result and target execution result of the second initial task in the second task topology. The second initial task in the second task topology of multiple task topologies is not affected by interactive operations. While updating the first initial task using operation information, multiple second initial tasks in the second task topology can continue to be executed using the large model. Feedback information is generated using the initial execution result and target execution result of the second initial task. This reduces the number of initial tasks updated in the task orchestration strategy, avoids redundant computational overhead and waiting time caused by a full adjustment of the task orchestration strategy required for the large model, and improves the response efficiency of providing feedback information to the target object, thereby enhancing the user experience.

[0081] In some embodiments, using a large model to perform semantic understanding on the operation information of interactive operations and updating the first initial task related to the operation information in the first task topology may include: using a large model to perform semantic understanding on the operation information and requirement information to obtain a semantic understanding result; and updating multiple first initial tasks in the first task topology based on the semantic understanding result.

[0082] According to embodiments of this disclosure, the semantic understanding result can represent the degree of semantic relevance between demand information and operational information. For example, the semantic understanding result can represent the semantic difference between operational information and demand information, changes in demand intent, potential demand intent, etc.

[0083] For example, a large model can perform semantic understanding on the demand information "basketball game result" and the interactive operation information "yesterday's A-League football game result," determining that the semantic understanding result represents "the target object is concerned about the basketball game result and also wants to know yesterday's A-League football game result." Therefore, based on the semantic understanding result, the change in the target object's demand intention can be represented as a change in demand from basketball game result to A-League football game result, and the potential demand intention can represent interest in multiple types of popular matches.

[0084] Therefore, the large model can use semantic understanding results as prompts to update the task attribute parameters of the first initial task in the first task topology. This enables the target task to call the retrieval tool to retrieve this week's basketball game results and yesterday's A-League football game results from authorized match result websites. Simultaneously, a football game result analysis task can be added as a target task to the first task topology. By utilizing the large model to execute the target task, analysis content can be output. This analysis content can focus on the impact of yesterday's multiple football game results on the league rankings of multiple league teams through textual analysis.

[0085] According to embodiments of this disclosure, a large model is used to perform semantic understanding on demand information and operational information, so that the semantic understanding result represents the degree of semantic correlation between the demand information and operational information. This allows the large model to synchronously update multiple first initial tasks in the same task topology based on the potential demand intents and changes in demand intents represented by the semantic understanding results, thereby improving the accuracy and efficiency of updating the initial tasks. Simultaneously, by synchronously executing the second initial task in the second task topology during the updating of the first task topology, redundant computational overhead and redundant waiting time caused by regenerating the task orchestration strategy can be avoided.

[0086] In some embodiments, using a large model to understand the intent of interactive operation information and update the first initial task related to the operation information in the first task topology may include: based on tool description prompts, using the first large model to understand the tool invocation intent of the operation information to obtain the tool invocation task in the first task topology.

[0087] According to embodiments of this disclosure, tool description prompts represent tool description information related to preset tools. Preset tools can be pre-defined service resources within the system, or external service resources that can be invoked under authorized conditions. The tool description information describes the preset tool's parameter range, applicable scenarios, output result types, and other information used to describe the preset tool's performance.

[0088] In some embodiments, the first major model can understand the tool invocation intent contained in the operation information based on the tool performance of the preset tool indicated by the tool description prompt, and generate a tool invocation task that matches the tool invocation intent as the target task to update the first initial task in the first task topology. This can make the first task topology in the updated task orchestration strategy more accurately adapt to the actual tool invocation needs of the target object, thereby improving the matching degree between the target execution result and the demand intent, and further improving the matching degree between the feedback information and the demand intent, thereby improving the user experience.

[0089] According to an embodiment of this disclosure, the tool invocation task instructs the second major model to invoke a target tool in a preset tool by executing the tool invocation task, to execute at least one specified task in the first task topology, and to obtain the target tool execution result for determining feedback information.

[0090] For example, a tool invocation task could be a code editing task. By utilizing the second major model to invoke a code editing tool as the target tool, the specified task is executed, and the output is a code script that matches the task attribute information of the tool invocation task. This code execution script serves as the execution result of the target tool. Thus, the second major model can be used to execute code execution tasks in the first task topology that are dependent on the code editing task. The second major model invokes a code execution tool as the target tool to execute the code script, and the resulting code execution is used as the execution result of the target tool.

[0091] According to embodiments of this disclosure, by using a first major model to update the initial task in the task orchestration strategy, and using a second major model to execute the tool invocation task as the target task to invoke the corresponding target tool to execute the specified task according to the task attribute information of the tool invocation task, multiple different major models can be used for task orchestration and tool invocation. In combination with the use of the second major model to synchronously execute the second initial task in the second task topology, the response efficiency for interactive operations can be improved, and the overall execution efficiency of the task orchestration strategy and the information quality of the target execution result can be improved. This will improve the matching degree between feedback information and the needs and intentions of the target object, and achieve the technical effect of improving user experience.

[0092] In some embodiments, updating the first initial task related to the operation information in the first task topology by using a large model to understand the intent of the operation information of the interactive operation may also include: based on the tool description prompts, using the first large model to understand the execution time intent of the operation information to obtain a timed invocation task.

[0093] According to embodiments of this disclosure, a timed invocation task is used to instruct the second major model to invoke a target tool in a preset tool to perform a specified task at a specified time or period that matches the execution time intent represented by the operation information.

[0094] In some embodiments, the execution time intent represented by the operation information can be determined based on key information content such as keywords "8 PM tonight" carried by the operation information, and the key information content is related to a specified time or a specified time period.

[0095] In some embodiments, the execution time intent can also be determined by semantic understanding of the operation information using the first large model. By utilizing the large model to understand the execution time intent implied in the operation information, the tool's invocation time attribute can be determined, and the target tool can be identified from a set of preset tools based on the tool's capabilities as indicated by the tool description information. This allows the first large model to generate a time-based invocation of the target tool at a specified moment or time period matching the specified time intent, enabling the target tool to perform a specified task, thus achieving a scheduled invocation task.

[0096] For example, the operation information could be "to calculate the match results between Team A and Team B and create a match points table." By using the first major model to understand the execution time intent of the operation information, we can determine that the execution time intent in the understanding result is "the match between Team A and Team B that started at 8 PM tonight, between 10 PM and 10:20 PM after the match ends." Therefore, the scheduled task can be defined as: between 10 PM and 10:20 PM tonight, calling the retrieval tool to obtain the match results information of Team A and Team B from the authorized information interface.

[0097] According to embodiments of this disclosure, by performing semantic understanding on the operation information and tool description information of interactive operations during the execution of the editing task strategy, the scheduled task can be determined. This allows for the invocation of tools adapted to the needs of the target object according to time requirements to obtain tool invocation results. As a result, the feedback information can more accurately meet the diverse real-time needs of the target object, avoiding the delay in obtaining feedback information generated by the interaction with the large language model after a specified time or period. Furthermore, the scheduled task invocation intelligently and automatically responds to the delayed needs of the target object to improve the user experience.

[0098] In some embodiments, a task orchestration strategy may indicate different processing modes for multiple initial tasks or multiple target tasks. The processing modes shown in Tables 1.1 and 1.2 exemplify the processing modes that a task orchestration strategy can represent.

[0099] Table 1.1

[0100]

[0101] Table 1.2

[0102]

[0103] It should be noted that the processing modes shown in Tables 1.1 and 1.2 are only used to illustrate the execution process or application scenario of the task orchestration strategy for large models, and are not intended to limit the specific scope of the task orchestration strategy provided in the embodiments of this disclosure.

[0104] In some embodiments, the information processing method based on the large model may further include: based on the operation information of the interactive operation, using the computing unit that deploys the large model to stop the execution of at least one initial task in the task orchestration strategy.

[0105] According to embodiments of this disclosure, the computing unit may include any type of device or component capable of performing computing tasks based on model parameters of a large model, such as a central processing unit (CPU) or a graphics processing unit (GPU).

[0106] In some embodiments, upon receiving operation information, the computation unit can be paused to execute the initial task according to the model parameters, and the initial task indicated by the interactive operation in the task orchestration strategy can be updated based on the task attribute information represented by the operation information to obtain the updated target task. After obtaining the target task, the computation unit is used to execute the target task according to the model parameters of the large model to obtain the target execution result. This avoids the computational overhead redundancy and excessive delay caused by re-orchestrating tasks based on the operation information and context after receiving the interactive operation information. By stopping the computation unit from executing the initial task and resuming the execution of the task orchestration strategy after updating the target task, the computational load of the computation unit can be flexibly controlled and computational overhead can be saved during the response to the interactive questions of the target object.

[0107] In some embodiments, the orchestration strategy includes multiple task topologies that are dependent on a specified initial task, and the stopped state topology among the multiple task topologies includes the initial task that has been stopped from execution.

[0108] According to embodiments of this disclosure, the information processing method based on a large model may further include: during the process of stopping the execution of at least one initial task in a task orchestration strategy using a computing unit that deploys the large model, executing an initial task in an associated topology using the computing unit.

[0109] According to embodiments of this disclosure, the associated topology is any of the multiple task topologies other than the stopped state topology.

[0110] For example, the associated topology can be the second task topology among multiple task topologies that has not yet been updated based on operational information.

[0111] For example, the associated topology can be the first task topology among multiple task topologies that has been updated in a historical period.

[0112] By utilizing the computing unit to stop executing the initial task in the stopped topology, redundant computational overhead caused by executing the initial task in the stopped state can be avoided. Furthermore, while the stopped topology is being stopped by the computing unit, at least one of the initial and target tasks in the associated topology, which are unrelated to the current interactive operation, can continue to be executed. This improves the execution efficiency of the task orchestration strategy, thereby enhancing the user experience by increasing the response speed of output feedback information.

[0113] Figure 4 The illustration shows a schematic diagram of the principle of a large-model-based information processing method according to an embodiment of the present disclosure.

[0114] like Figure 4As shown, the task orchestration strategy in this embodiment can be a manuscript task orchestration strategy 400 for generating competition news releases. The manuscript task orchestration strategy 400 can include multiple initial tasks, which can be based on... Figure 4 The circular elements shown represent the initial tasks. These are designated as Initial Task 1 through Initial Task 6. Initial Task 1 represents the manuscript outline generation task, Initial Task 2 represents the match image material retrieval task, Initial Task 3 represents the match image material editing task, Initial Task 4 represents the match process text retrieval task, Initial Task 5 represents the player data retrieval task, and Initial Task 6 represents the manuscript semantic fusion task.

[0115] In this scenario, when the target object performs an interactive operation on the fourth initial task and inputs operation information, the second computing unit stops executing the fourth initial task. The task topology corresponding to the fourth initial task can be the stopped task topology 420. During the process of updating the fourth initial task based on the operation information, the first computing unit can continue to execute the initial task in the associated topology 410 to retrieve and edit the competition image materials in real time. This allows for dynamic stopping of the calculation process for the task topology based on multiple different computing units, improving the execution efficiency of the manuscript task arrangement strategy 400. It also avoids the additional computational overhead of re-arranging tasks and improves the user experience by reducing computational overhead and feedback latency.

[0116] In some embodiments, during the execution of a task orchestration strategy using a large model, updating at least one initial task in the task orchestration strategy based on the interaction operation of the target object may further include: during the execution of a task orchestration strategy using a large model, updating the initial execution result of the initial task based on the operation information carried by the interaction operation for the initial task, thereby obtaining a first target execution result related to the initial task.

[0117] According to embodiments of this disclosure, the execution result of the first target is used as input data for at least one initial task or target task that has not been completed in the orchestration strategy.

[0118] For example, the initial execution result could be "match result data between Team A and Team B". If the match result between Team A and Team B is verified and revised by the referee after the match, the match result data changes. The target object can perform interactive operations on the revised final match result to modify the match score data in the initial execution result "match result data between Team A and Team B", thus obtaining the modified final score data as the first target execution result. This final score data is input into another initial task that depends on the initial task: "Analyze the match between Team A and Team B". Thus, a large model can be used to process the "final score data" to output the match analysis text.

[0119] According to embodiments of this disclosure, during the execution of a task orchestration strategy, the target object can modify the initial execution result of the initial task in real time and apply the modified target execution result to subsequent initial or target tasks that have not yet been completed. This allows the target object to obtain timely execution results that meet its needs through interactive operations during the execution of the orchestration strategy. Furthermore, the input data of subsequent initial or target tasks can be adjusted based on the target execution result generated from the operation information, thereby adjusting the execution result of the orchestration strategy according to the target object's needs. This improves the matching degree between the target execution result and the target object's needs and intentions, and enhances the efficiency of feedback information generation in a timely and convenient manner. It avoids redundant computational overhead and waiting time caused by repeatedly generating task orchestration strategies, thus satisfying the actual needs and intentions of the target object with high quality and improving the user experience.

[0120] In some implementations, during the execution of a task orchestration strategy using a large model, updating at least one initial task in the task orchestration strategy based on the interactive operations of the target object may also include: using the large model to plan tasks based on the execution results and requirements information of the first target, to obtain target tasks that are dependent on at least one initial task that has not been completed.

[0121] In one embodiment, a first-level model can be used to process the demand information "generate a press release for the match between Team A and Team B" and the first objective execution result "2:1" (the modified final score). The first-level model can then be used to understand the intent of the demand information and the first objective execution result, resulting in the interaction intent "the match has been reversed, and a press release needs to be generated based on the new score and the reversal process." The first-level model can then generate the objective task "retrieve real-time match commentary text for the match between Team A and Team B" by processing the interaction intent. This objective task can then be added to the task orchestration strategy, and a second-level model can be used to execute the second objective task to obtain the second objective execution result "real-time match commentary text." Finally, based on the second-level model, a press release reporting on the reversed match is obtained by semantically fusing the second objective execution result with the modified first objective execution result.

[0122] According to embodiments of this disclosure, task planning using a large model based on the execution result of a first target and requirement information can include using the first large model to understand the intent of the target object performing interactive operations based on the first target execution result generated from operation information, requirement information, and other contextual content, to determine the actual interactive intent of the target object performing the interactive operations. This allows task planning based on the actual interactive intent generated by the target object through performing interactive operations, generating target tasks that satisfy the actual interactive intent. The modified first target execution result can then be used as input data into the target task, enabling the second large model to execute the target task based on the first target execution result. This ensures that the target execution result satisfies the actual interactive intent of the target object and improves the matching degree between feedback information and the actual requirement intent of the target object. This achieves timely satisfaction of the target object's needs based on dynamic adjustments to the task orchestration strategy, improves the data quality of feedback information, and enhances the user experience.

[0123] In some embodiments, determining feedback information that matches the target object's needs and intentions based on the target execution results determined by using a large model to perform target task execution may include: using a large model to perform semantic fusion on multiple target execution results based on style attributes related to the needs information to obtain feedback information.

[0124] According to embodiments of this disclosure, style attributes can represent the linguistic expression style of text or copywriting, the image style attribute of an image, the pronunciation style attribute of audio data, etc. Style attributes can be the intent content within the demand intent determined by intent understanding of demand information using a large model, or style attributes can also be the information content carried within the demand information. Embodiments of this disclosure do not limit the specific type of style attribute or the specific method of determining the style attribute.

[0125] In one embodiment, semantic fusion of the execution results of multiple targets using a large model may include semantic fusion of multiple text fragments based on prompt words determined by style attributes, to obtain feedback information expressed based on style attributes.

[0126] In one embodiment, the style attribute can be the information content in the operation information carried by the interactive operation. The target object inputs the style attribute through the interactive operation, and can adjust the style attribute of the feedback information in real time based on actual needs. This allows the target object to modify the style attribute of the feedback information and the quality of the feedback information through real-time interaction with the large model, thereby improving the matching degree of the style attribute of the feedback information to the specific scenario.

[0127] In some embodiments, style attributes can be associated with a specified demand scenario. For example, a style attribute can be "the language expression style for reporting and explaining to middle school students". Based on the style cue words determined by the style attribute, the large model can be prompted to perform stylized semantic fusion on the speech fragments output by multiple target tasks, so that the resulting speech can meet the expression style of the demand scenario of "reporting and explaining to middle school students", and make the feedback information match the scenario requirements of the target object.

[0128] Figure 5 A schematic diagram of the system architecture of an information processing system according to an embodiment of the present disclosure is shown.

[0129] like Figure 5 As shown, the information processing system 500 can apply the large-model-based information processing method provided in this embodiment. The information processing system 500 may include a model service layer, a security risk control layer, an intelligent scheduling layer, a message communication layer, a tool service layer, a multimodal rendering layer, a knowledge enhancement layer, a file management layer, and a memory layer.

[0130] The model service layer can deploy multiple large models to execute the large model-based information processing method provided according to embodiments of this disclosure. For example, the model service layer can provide an inference large model, a scheduling large model, an execution large model, and a reflection large model. The inference large model and the scheduling large model can serve as the first large model for task orchestration and initial task updates, while the execution large model can serve as the second large model for executing the initial or target task. The reflection large model can examine the initial or target execution results output by each of the second large models and output reflection results. Based on these reflection results, the second large model can be iteratively prompted to re-execute the initial or target task to output high-quality execution results that meet the requirements of the target object.

[0131] The task orchestration strategy output by the inference big model can represent the dependencies between multiple initial tasks using topological forms such as Chain-of-Thought (CoT), Tree of Thoughts (ToT), and Graph of Thoughts (GoT). The scheduling big model can select the optimal second big model from a pre-set big model library to execute the initial or target task based on a model routing mechanism. For example, the second big model can be selected based on the latency, computational cost, and accuracy metrics of the pre-set big model. The reflection big model establishes a multi-dimensional error type classification system, constructs an anomaly detection matrix based on quality evaluation options such as logical contradictions, factual errors, and task execution anomalies, evaluates the quality of the execution results, and uses the reflection results as input to obtain the second big model, thereby controlling the output of the second big model to achieve better execution results. This enables a closed-loop execution process of reflection-correction-verification for the task orchestration strategy. Furthermore, the reflection big model can also perform reflection checks on the execution results based on public or private knowledge bases in the knowledge enhancement layer to improve the accuracy of the reflection results.

[0132] The scheduling master model can select parallel or serial processing modes to schedule the second master model to execute the task orchestration strategy based on the dependencies indicated by the task orchestration strategy. Simultaneously, the scheduling master model can understand tool description information and provide tool resources to the execution master model based on the model context protocol. The execution master model can select the target tool based on multiple tool descriptions and invoke the target tool to execute the initial or target task through the model context protocol, thereby improving the accuracy of tool invocation. Furthermore, the execution master model can perform stylistic fusion of the execution results based on style attributes, ensuring that the style attributes of the feedback information determined according to the target execution results meet the needs of the target object.

[0133] The intelligent scheduling layer can support various scheduling modes, such as parallel, asynchronous, and serial modes, for large-scale task outputs through a multi-mode scheduling engine. It can also allocate computing resources across multiple computing units or devices based on load balancing mechanisms to meet the efficiency requirements of task orchestration strategies. The task coordination mechanism module of the intelligent scheduling layer can also be used for monitoring and managing the entire lifecycle of initial and target tasks, including creation, execution, pause, and resumption, and supports features such as resuming task execution from breakpoints.

[0134] The message communication layer can enable message communication between different large models and between models and tools by building standardized communication protocols. It also categorizes and stores messages such as operation information from the target object's interactive operations, execution results output by the large model, and the execution status of the initial or target task.

[0135] The file management layer can parse and store multimodal information such as speech, text, and images through a multimodal processing module. For example, it can extract semantic features from the context using sliding windows and attention mechanisms, providing input data for large models during task updates or task orchestration. The file management layer can also support information retrieval based on scalar or vector indexes. The hybrid storage module can use hierarchical storage to store structured information as scalars, semantic information as vectors, and raw information as objects. Furthermore, it can record data generated by large models during execution through log files.

[0136] The tool service layer can standardize the communication between internal tools and external third-party tool resources through the model context protocol module, supporting protocol-level extensibility to facilitate interactive data output, large model or agent access, and other functions. Different internal tools and external third-party tool resource interfaces can be configured in the preset tool library for system invocation.

[0137] The tool service layer also allows for visual configuration of workflows, supporting intelligent switching between serial and parallel modes, retry, circuit breaking, and configuration of degradation exception handling strategies. It can also monitor metrics such as tool call success rate, tool response latency, and resource consumption, while comprehensively recording tool inputs, outputs, and intermediate states.

[0138] The entire memory layer can store multi-source data such as user behavior, dialogue context, and personalized preference attributes based on graph nodes of a graph neural network. It supports unified storage and associated retrieval of text, files, and structured data to achieve memory storage, and can combine scalar metadata indexes and vector semantic indexes to retrieve stored data.

[0139] The knowledge base content in the knowledge enhancement layer supports the parsing and storage of various file formats, including text, images, documents, audio, and video. It supports file format normalization and multi-dimensional scalar and vector retrieval. The public domain knowledge base integrates high-quality document library content, real-time internet data, image search, and multimedia resources. The private domain knowledge base supports scenario-based knowledge management, including document repositories, personal knowledge spaces, and enterprise cloud storage.

[0140] The security risk control layer supports a multi-level defense system to perform multiple checks on input information to intercept sensitive information, and conducts content security reviews of output content based on keywords and risk control classification models. Furthermore, it can combine manual review to intercept abnormally sensitive information. The security risk control layer also supports diverse functions such as privacy data compliance, access control systems, and review process tracking. The multimodal rendering layer can realize real-time compilation, transcoding, or playback of various files.

[0141] Figure 6A flowchart illustrating a large-model-based information processing method according to another embodiment of this disclosure is shown.

[0142] like Figure 6 As shown, the information processing method based on a large model in this embodiment can execute steps S601 to S606 based on an information processing system.

[0143] In step S601, the requirement information can be received, and the first major model in the model service layer can be called to execute step S602. Based on the requirement information, the thinking process is performed to determine the requirement intent of the target object.

[0144] In step S603, the first large model is used to plan the task orchestration strategy, or the initial task is updated based on the interaction of the target object.

[0145] In step S604, the initial task and the target task are executed using the scheduling strategy provided by the intelligent scheduling layer, such as the second major model, to obtain the initial execution result and the target execution result. For example, the second major model can be used to call preset tools in the tool service layer to execute tool calling tasks.

[0146] In step S605, the reflective large model in the model layer is used to perform a self-check on the initial execution result and the target execution result. For example, the execution result can be checked based on the knowledge base content provided by the knowledge enhancement layer to output a reflective result. If the reflective result in step S605 indicates that the self-check has passed, feedback information can be obtained by fusing multiple target execution results. Then, step S606 is executed to render the feedback information based on the multimodal rendering layer. During the execution of the information processing method, the security risk control layer can detect the security risks of the output results and interactive operation information in real time to avoid outputting sensitive or non-compliant information.

[0147] Figure 7 A block diagram of a large-model-based information processing apparatus according to an embodiment of the present disclosure is shown schematically.

[0148] like Figure 7 As shown, the information processing device 700 based on a large model includes: a first acquisition module 710, a second acquisition module 720, and a feedback information determination module 730.

[0149] The first acquisition module 710 is used to perform task orchestration on the requirement information related to the target object to obtain a task orchestration strategy containing multiple initial tasks.

[0150] The second acquisition module 720 is used to update at least one initial task in the task orchestration strategy based on the interaction operation of the target object during the execution of the task orchestration strategy using a large model, so as to obtain the target task.

[0151] The feedback information determination module 730 is used to determine feedback information that matches the needs and intentions of the target object based on the target execution results determined by using the large model to execute the target task.

[0152] According to embodiments of this disclosure, the orchestration strategy includes multiple task topologies that are dependent on a specified initial task, and the interaction operation is related to a first task topology among the multiple task topologies; wherein, the second obtaining module includes a first updating unit.

[0153] The first update unit is used to update the first initial task related to the operation information in the first task topology by using the large model to understand the intent of the operation information of the interaction operation. The feedback information is determined based on the initial execution result and the target execution result of the second initial task in the second task topology.

[0154] According to embodiments of this disclosure, the first update unit includes a semantic understanding subunit and an update subunit.

[0155] The semantic understanding subunit is used to leverage the large model to perform semantic understanding on operational and demand information, and obtain semantic understanding results that represent the degree of semantic correlation between demand and operational information.

[0156] The update sub-unit is used to update multiple first initial tasks in the first task topology based on the semantic understanding results.

[0157] According to embodiments of this disclosure, the first update unit includes a tool invocation task acquisition subunit.

[0158] The tool invocation task acquisition sub-unit is used to understand the tool invocation intent of the operation information based on the tool description prompt words and the first major model to obtain the tool invocation task in the first task topology. The tool description prompt words represent the tool description information related to the preset tool. The tool invocation task instructs the second major model to invoke the target tool in the preset tools by executing the tool invocation task, execute at least one specified task in the first task topology, and obtain the target tool execution result used to determine the feedback information.

[0159] According to embodiments of this disclosure, the first update unit includes a task acquisition subunit that is periodically invoked.

[0160] The scheduled task acquisition sub-unit is used to understand the execution time intent of the operation information based on the tool description prompt words and the first major model to obtain the scheduled task. The tool description prompt words represent the tool description information related to the preset tool. The scheduled task is used to instruct the second major model to call the target tool in the preset tools to execute the specified task at a specified time or time period that matches the execution time intent represented by the operation information.

[0161] According to embodiments of this disclosure, the information processing apparatus based on a large model further includes a stop execution module.

[0162] The stop execution module is used to stop at least one initial task in the task orchestration strategy based on the operation information of the interactive operation and the computing unit of the deployed large model.

[0163] According to embodiments of this disclosure, the orchestration strategy includes multiple task topologies that have dependencies on a specified initial task, and the stopped state topology among the multiple task topologies includes the initial task that has been stopped from execution. The information processing device based on the large model further includes an execution module.

[0164] The execution module is used to execute initial tasks in associated topologies during the process of stopping at least one initial task in the task orchestration strategy using the computing unit that deploys the large model. The associated topologies are other task topologies among multiple task topologies other than the stopped state topology.

[0165] According to embodiments of this disclosure, the second obtaining module includes a second updating unit.

[0166] The second update unit is used to update the initial execution result of the initial task based on the operation information carried by the interaction operation for the initial task during the execution of the task orchestration strategy using the large model, so as to obtain the first target execution result related to the initial task. The first target execution result is used as input data for at least one initial task or target task that has not been completed in the orchestration strategy.

[0167] According to embodiments of this disclosure, the second obtaining module further includes a target task obtaining unit.

[0168] The target task acquisition unit is used to perform task planning based on the execution results and requirements of the first target using the large model, and to obtain target tasks that are dependent on at least one initial task that has not been completed.

[0169] According to embodiments of this disclosure, the first obtaining module includes: a demand intent obtaining unit and a task orchestration strategy obtaining unit.

[0170] The demand intent acquisition unit is used to perform semantic understanding of demand information using a large model to obtain demand intent.

[0171] The task orchestration strategy acquisition unit is used to orchestrate tasks based on demand intent and obtain task orchestration strategies.

[0172] According to embodiments of this disclosure, the demand intent includes a desired time intent that characterizes the target object's expectation of obtaining feedback information; the task orchestration strategy acquisition unit includes a task orchestration strategy acquisition subunit.

[0173] The task orchestration strategy acquisition sub-unit is used to determine the delay prompt information based on the expected time intent. The first major model is used to orchestrate the demand content of the demand intent to obtain the task orchestration strategy. The second major model executes the task orchestration strategy and obtains feedback information. The strategy execution time is matched with the expected time intent.

[0174] According to embodiments of this disclosure, the large-model-based information processing apparatus further includes a large-model determination module.

[0175] The large model determination module is used to determine the second large model for performing the initial task based on the task attribute information of the initial task, using the first large model to perform semantic understanding based on the model description information of multiple candidate large models, and the model description information characterizes the model performance attributes of the candidate large models.

[0176] According to embodiments of this disclosure, the demand intent obtaining unit includes a first processing subunit.

[0177] The first processing subunit is used to perform semantic understanding of demand information based on the object attribute knowledge graph and a large model. The object attribute nodes in the object attribute knowledge graph represent the historical interaction information or object preference attributes of the target object.

[0178] According to embodiments of this disclosure, the feedback information determination module includes:

[0179] The feedback information acquisition unit is used to obtain feedback information by semantically fusing the execution results of multiple targets using a large model based on style attributes related to the demand information.

[0180] According to embodiments of this disclosure, the information processing apparatus based on a large model further includes a push module.

[0181] The push module is used to push at least one of the task orchestration strategy and the initial execution result of the initial task to the target object.

[0182] Figure 8 A schematic block diagram of an artificial intelligence agent according to an embodiment of the present disclosure is shown.

[0183] In embodiments of this disclosure, such as Figure 8 As shown, the AI ​​agent 800 may include an input module 810, a processing module 820, and an output module 830.

[0184] Input module 810 is used to receive input information;

[0185] The processing module 820 is used to determine the target task based on the input information received by the input module, determine the large model based on the target task, and obtain output information by calling the large model to execute the information processing method based on the large model according to the embodiments of this disclosure.

[0186] Output module 830 is used to output the output information obtained by the processing module.

[0187] According to embodiments of this disclosure, the input module 810 is responsible for receiving or sensing information such as queries, requests, instructions, signals, or data from the outside world (e.g., users or the external environment), and converting it into a format that the AI ​​agent 800 can understand and process. The input module 810 is the primary link for the AI ​​agent 800 to interact with the outside world, enabling the AI ​​agent 800 to efficiently and accurately obtain necessary "sensory" information from the outside world and respond to this information.

[0188] In the example, the input module 810 can input the requirement information and interactive operation information described above.

[0189] In the example, processing module 820 is the core support for the AI ​​agent 800's ability to handle complex tasks. Processing module 820 can execute the large-model-based information processing methods described above.

[0190] In the example, the performance of the processing module 820 is closely related to the large model on which the AI ​​agent 800 is based. To fully leverage the capabilities of the large model, the internal structure of the processing module 820 can be designed to be highly configurable and scalable to handle various types of tasks and requirements in real-world scenarios.

[0191] In the example, after the AI ​​agent 800 acquires the required voice, the processing module 820 can use a large model to process the required information to obtain a task orchestration strategy. The large model executes the task orchestration strategy and updates the initial task based on the interactive operations to obtain the target task. Feedback information is obtained by processing the target execution result using the large model, and this feedback information is then passed to the output module 830.

[0192] Understandably, while large language models possess excellent language understanding and generation capabilities, like humans, their ability to solve tasks is limited without the aid of any tools. However, once the AI ​​agent 800 is given the ability to invoke tools, it can perform tasks such as using a calculator to complete mathematical calculations, using Python to perform data analysis, and using a search engine to create weather forecasts.

[0193] In the example, output module 830 can output the feedback information described above.

[0194] The AI ​​agent 800 according to embodiments of this disclosure can simply and effectively improve the level of intelligence, as well as enhance flexibility and versatility.

[0195] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0196] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0197] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the method described above.

[0198] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0199] Figure 9 A schematic block diagram of an example electronic device for implementing the large-scale model-based information processing method of embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0200] like Figure 9 As shown, device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 902 or a computer program loaded from storage unit 908 into random access memory (RAM) 903. RAM 903 may also store various programs and data required for the operation of device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0201] Multiple components in device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of monitors, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0202] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as large model-based information processing methods. For example, in some embodiments, the large model-based information processing method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the large model-based information processing method described above can be performed. Alternatively, in other embodiments, the computing unit 901 can be configured to perform large model-based information processing methods by any other suitable means (e.g., by means of firmware).

[0203] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0204] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0205] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0206] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0207] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0208] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0209] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0210] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information processing method based on a large model, comprising: Task orchestration is performed on the requirements information related to the target object to obtain a task orchestration strategy containing multiple initial tasks; During the execution of the task orchestration strategy using the second large model, based on the interactive operation input by the target object, at least one of the initial tasks in the task orchestration strategy is updated using the first large model to obtain the target task. Based on the operation information of the interactive operation, the execution of at least one initial task in the task orchestration strategy is stopped using the computing unit that deploys the second large model. The orchestration strategy includes multiple task topologies that have a dependency relationship with the specified initial task. The stopped state topology among the multiple task topologies includes the initial task that has been stopped from execution. The stopped initial task is updated using the first large model based on the operation information of the interactive operation. During the process of stopping the execution of the initial task in the stopped state topology using the computing unit that deploys the second large model, the initial task in the associated topology that is unrelated to the operation information of the interactive operation is executed using the computing unit. The associated topology is the other task topologies among the multiple task topologies besides the stopped state topology. as well as Based on the initial execution results of the initial task in the associated topology and the target execution results determined by executing the target task using the second major model, feedback information matching the needs and intentions of the target object is determined.

2. The method according to claim 1, wherein, The orchestration strategy includes multiple task topologies that are dependent on a specified initial task, and the interaction operation is related to a first task topology among the multiple task topologies; The step of updating at least one initial task in the task orchestration strategy based on the interaction operation of the target object using the first large model includes: By using the first large model to understand the intent of the interactive operation information, the first initial task related to the operation information in the first task topology is updated to obtain the target task. The task orchestration strategy includes the first task topology and the second task topology. The feedback information is determined based on the initial execution result of the second initial task in the second task topology and the target execution result. The associated topology includes the second task topology.

3. The method according to claim 2, wherein, The step of using the first large model to understand the intent of the interactive operation and updating the first initial task related to the operation information in the first task topology includes: By using the large model to perform semantic understanding on the operation information and the demand information, a semantic understanding result is obtained, which represents the degree of semantic correlation between the demand information and the operation information. The first initial tasks in the first task topology are updated based on the semantic understanding results.

4. The method according to claim 2, wherein, The step of using the first large model to understand the intent of the interactive operation and updating the first initial task related to the operation information in the first task topology includes: Based on the tool description prompts, the first large model is used to understand the tool invocation intent of the operation information to obtain the tool invocation task in the first task topology. The tool description prompts represent tool description information related to preset tools. The tool invocation task instructs the second large model to invoke the target tool in the preset tools by executing the tool invocation task, execute at least one specified task in the first task topology, and obtain the target tool execution result used to determine the feedback information.

5. The method according to claim 2, wherein, The step of using the first large model to understand the intent of the interactive operation and updating the first initial task related to the operation information in the first task topology includes: Based on the tool description prompts, the first major model is used to understand the execution time intent of the operation information to obtain a timed invocation task. The tool description prompts represent tool description information related to preset tools, and the timed invocation task is used to instruct the second major model to invoke the target tool in the preset tools to execute the specified task at a specified time or time period that matches the execution time intent represented by the operation information.

6. The method according to claim 1, wherein, In the process of executing the task orchestration strategy using the second large model, updating at least one of the initial tasks in the task orchestration strategy using the first large model based on the interactive operations of the target object includes: During the execution of the task orchestration strategy using the second major model, based on the operation information carried by the interactive operation for the initial task, the initial execution result of the initial task is updated using the first major model to obtain a first target execution result related to the initial task. The first target execution result is used as input data for at least one initial task or target task that has not been completed in the orchestration strategy.

7. The method according to claim 6, wherein, The step of updating at least one of the initial tasks in the task orchestration strategy based on the interaction operations of the target object during the execution of the task orchestration strategy using the second large model further includes: Using the large model, task planning is performed based on the execution result of the first objective and the requirement information to obtain target tasks that are dependent on at least one of the initial tasks that have not been completed.

8. The method according to claim 1, wherein, The task orchestration strategy, which involves arranging requirements information related to the target object to obtain a task orchestration strategy containing multiple initial tasks, includes: The first major model is used to perform semantic understanding on the demand information to obtain the demand intent; and Based on the stated requirements, tasks are orchestrated to obtain the task orchestration strategy.

9. The method according to claim 8, wherein, The demand intent includes the expected time intent that characterizes the target object's expectation of obtaining the feedback information; The step of orchestrating tasks based on the stated demand intent to obtain the task orchestration strategy includes: Based on the delay prompt information determined by the expected time intention, the first major model is used to orchestrate the demand content of the demand intention to obtain the task orchestration strategy. The execution time of the second major model to execute the task orchestration strategy and obtain the feedback information is matched with the expected time intention.

10. The method according to claim 8 or 9, wherein, The method further includes: Based on the task attribute information of the initial task, the first large model performs semantic understanding based on the model description information of multiple candidate large models, and determines a second large model from the multiple candidate large models to perform the initial task. The model description information characterizes the model performance attributes of the candidate large models.

11. The method according to claim 8, wherein, The semantic understanding of the demand information using the first major model includes: Based on the object attribute knowledge graph, the large model is used to perform semantic understanding of the demand information, wherein the object attribute nodes in the object attribute knowledge graph represent the historical interaction information or object preference attributes of the target object.

12. The method according to claim 1, wherein, The feedback information is determined based on the following operations: Based on the style attributes related to the required information, the second major model is used to perform semantic fusion on the execution results of multiple targets to obtain the feedback information.

13. The method according to claim 1, wherein, The method further includes: Push at least one of the task orchestration strategy and the initial execution result of the initial task to the target object.

14. An information processing device based on a large model, comprising: The first acquisition module is used to orchestrate the requirements information related to the target object and obtain a task orchestration strategy containing multiple initial tasks. The second acquisition module is used to, during the execution of a task orchestration strategy using the second large model, update at least one of the initial tasks in the task orchestration strategy using the first large model based on the interactive operation input by the target object to obtain the target task; based on the operation information of the interactive operation, stop the execution of at least one initial task in the task orchestration strategy using the computing unit that deploys the second large model; the orchestration strategy includes multiple task topologies that have a dependency relationship with the specified initial task; the stopped state topology among the multiple task topologies includes the initial task that has been stopped from execution; update the stopped initial task using the first large model based on the operation information of the interactive operation; during the process of stopping the execution of the initial task in the stopped state topology using the computing unit that deploys the large model, execute the initial task in the associated topology that is unrelated to the operation information of the interactive operation using the computing unit; the associated topology is the other task topologies among the multiple task topologies besides the stopped state topology. as well as The feedback information determination module is used to determine feedback information that matches the needs and intentions of the target object based on the initial execution results of the initial task in the associated topology and the target execution results determined by executing the target task using the second major model.

15. An intelligent agent of artificial intelligence, comprising: The input module is used to receive input information; The processing module is configured to determine a target task based on the input information received by the input module, determine a large model based on the target task, and execute the method of any one of claims 1 to 13 by calling the large model to obtain output information; An output module is used to output the output information obtained by the processing module.

16. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 13.

17. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 13.

18. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 13.

Images