Task processing method, system and related device

Abstract

The invention discloses a task processing method and system and a related device. The method comprises the steps of obtaining a task processing instruction triggered by at least one target task; inputting the task processing instruction into an intelligent analysis model to obtain processing content which is output by the intelligent analysis model and matched with each target task; wherein the intelligent analysis model comprises a model basic network and at least one target adjustment sub-network parallel to the model basic network, and the model basic network is coupled with a scheduling module. And the scheduling module is used for selecting a target parameter set matched with a target task from a plurality of trained candidate parameter sets so as to construct the target adjustment sub-network. In this way, the task processing efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a task processing method, system and related devices. Background Art

[0002] With the continuous development of intelligence, the use of large language models for task processing is being applied in more and more fields. In order to improve the processing performance of large language models, large language models are usually fine-tuned for tasks of different types and fields, and all fine-tuning parameters generated during the fine-tuning process are added to the large language model, resulting in the operation of large language models consuming a large amount of computing resources and storage resources.

[0003] In view of this, how to reduce the computational and storage overhead of the model while maintaining high performance has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a task processing method, system and related devices, which can improve the efficiency of task processing.

[0005] In order to solve the above technical problems, a technical solution adopted in the present application is: providing a task processing method, comprising: obtaining a task processing instruction triggered by at least one target task; inputting the task processing instruction into an intelligent analysis model to obtain processing content output by the intelligent analysis model that matches each target task; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to a scheduling module, and the scheduling module is used to select a target parameter set that matches the target task from multiple trained candidate parameter sets to construct the target adjustment sub-network.

[0006] To solve the above technical problems, another technical solution adopted in the present application is: to provide a task processing system, including: an acquisition module, used to acquire task processing instructions triggered by at least one target task; a processing module, used to input the task processing instructions into an intelligent analysis model, and obtain the processing content output by the intelligent analysis model that matches each target task; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to a scheduling module, and the scheduling module is used to select a target parameter set that matches the target task from multiple trained candidate parameter sets to construct the target adjustment sub-network.

[0007] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, comprising: a memory and a processor coupled to each other, the memory storing program instructions, and the processor being used to execute the program instructions to implement the task processing method mentioned in the above technical solution.

[0008] In order to solve the above technical problems, another technical solution adopted in the present application is: providing a computer-readable storage medium on which program instructions are stored, and when the program instructions are executed by a processor, the task processing method mentioned in the above technical solution is performed.

[0009] The beneficial effect of the present application is that, different from the prior art, the task processing method proposed in the present application triggers a task processing instruction according to at least one acquired target task, and inputs the task processing instruction into the intelligent analysis model, so that the intelligent analysis model finally outputs the processing content matched by each target task at the same time, thereby improving the efficiency of task processing. In addition, the scheduling module coupled to the model basic network in the intelligent analysis model selects a target parameter set that matches the target task from a plurality of trained candidate parameter sets to construct a target adjustment subnetwork that is parallel to at least part of the analysis nodes in the model basic network, and the target adjustment subnetwork is used to adjust the output content of the matching analysis nodes, thereby improving the accuracy of the generated processing content. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0011] Figure 1 It is a flowchart of an implementation method of the task processing method of the present application;

[0012] Figure 2 yes Figure 1 Step S102 corresponds to a flow chart of another implementation method;

[0013] Figure 3 yes Figure 2 Step S201 corresponds to a flow chart of another implementation method;

[0014] Figure 4 yes Figure 2 Step S202 corresponds to a flow chart of another implementation method;

[0015] Figure 5 is a schematic diagram of a specific implementation scenario corresponding to step S402;

[0016] Figure 6 It is a structural diagram of an implementation method of the task processing system of the present application;

[0017] Figure 7 It is a structural schematic diagram of an embodiment of the electronic device of the present application;

[0018] Figure 8 It is a structural schematic diagram of an implementation method of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments, and different embodiments can be adaptively combined. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] The task processing method proposed in this application relies on an application on a smart terminal or at least a smart terminal with integrated task processing function, wherein the smart terminal can be a smart office notebook, a mobile phone, a tablet computer, a personal computer or a wearable smart device.

[0021] See also Figure 1 , Figure 1 It is a flowchart of an implementation method of the task processing method of the present application, and the method comprises:

[0022] S101: Acquire a task processing instruction triggered by at least one target task.

[0023] In one implementation, a plurality of target tasks are acquired, and task processing instructions are generated according to the plurality of target tasks.

[0024] Specifically, multiple target tasks are spliced ​​together, and separation marks are set between different target tasks to obtain task processing instructions. The target tasks include at least one of questioning, sentiment analysis, translation, and text generation.

[0025] In another embodiment, in response to the pre-setting of multiple prompt templates, corresponding content is extracted from each target task to fill in the corresponding area in the prompt template to obtain the prompt text corresponding to the target task, so as to facilitate the subsequent intelligent analysis model to understand the target task. After splicing the prompt texts corresponding to all target tasks, the task processing instruction is obtained.

[0026] S102: Input the task processing instruction into the intelligent analysis model to obtain the processing content matched with each target task output by the intelligent analysis model; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to a scheduling module, and the scheduling module is used to select a target parameter set matching the target task from a plurality of trained candidate parameter sets to construct a target adjustment sub-network.

[0027] In one embodiment, the task processing instructions including the target tasks are input into the intelligent analysis model, so that the model basic network and the target adjustment sub-network in the intelligent analysis model analyze the task processing instructions and generate processing content matching each target task.

[0028] Specifically, the scheduling module selects a target parameter set that matches the target task from multiple trained candidate parameter sets to construct a target adjustment subnetwork. By inputting the task processing instructions into the model base network in the intelligent analysis model, the input task processing instructions are analyzed and processed using the model base network, and the analysis and processing process of the model base network is adjusted using the target adjustment subnetwork, and finally the processing content that matches each target task is accurately output.

[0029] In one implementation scenario, the above-mentioned candidate parameter sets match the preset task identifiers, and each candidate parameter set is obtained by training using multiple training samples under the corresponding task identifier. The task identifier can be a task type, such as question answering, translation, and text generation, etc.; or, the task identifier can also be a task field, such as: medicine, mechanics, and computers, etc. For example, when the task identifier is medicine, in order to optimize the ability of the model base network to process different tasks in the medical field, multiple training samples in the medical field are obtained, and the trained model base network is fine-tuned using the LoRA algorithm to obtain the corresponding candidate parameter set while keeping the parameters in the model base network fixed, so that when processing medical-related tasks in the future, the model base network and the corresponding candidate parameter set are combined to generate more accurate processing content.

[0030] It should be noted that the model basic network includes multiple analysis nodes arranged in a stacked manner, at least some of the analysis nodes are arranged in parallel with target regulation sub-networks, and the target regulation sub-networks matched by the analysis node may be one or more. Among them, when the same analysis node is matched with multiple target regulation sub-networks, the multiple target regulation sub-networks matched by the analysis node are arranged in a stacked manner.

[0031] In addition, the model base network in the above intelligent analysis model is a large language model with data analysis capabilities, which is obtained through training. The task processing instructions containing the target task are input into the model base network so that the model base network understands the target task and outputs the corresponding processing content.

[0032] In a specific application scenario, the above-mentioned large language model may include but is not limited to deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory networks (LSTM), and generative pre-trained Transformer models, etc., and no specific restrictions are made on the specific structure and specific deployment of the large language model. In addition, it should be noted that the specific structure and specific deployment of the intelligent analysis model mentioned in other implementations of this application can refer to this implementation.

[0033] The task processing method proposed in the present application triggers a task processing instruction according to at least one acquired target task, and inputs the task processing instruction into the intelligent analysis model, so that the intelligent analysis model finally outputs the processing content matching each target task at the same time, thereby improving the efficiency of task processing. In addition, the scheduling module coupled to the model base network in the intelligent analysis model selects a target parameter set matching the target task from multiple trained candidate parameter sets to construct a target adjustment subnetwork parallel to at least part of the analysis nodes in the model base network. The target adjustment subnetwork is used to adjust the output content of the matching analysis nodes, which improves the accuracy of processing content generation while saving the storage consumption of the intelligent analysis model.

[0034] See also Figure 2 , Figure 2 yes Figure 1 Step S102 corresponds to a flowchart of another implementation method. In response to the task processing instruction being determined based on the target task input by multiple users, and each user being matched with corresponding user information, the specific implementation process of step S102 includes:

[0035] S201: Using the scheduling module to select a candidate parameter set matching the user information as a target parameter set, and constructing a target adjustment subnetwork according to the target parameter set.

[0036] In one embodiment, after the task processing instruction is obtained, according to the user information of the user corresponding to each target task in the task processing instruction, a scheduling module is used to select a target parameter set matching the user information from the candidate parameter set to construct a target adjustment subnetwork.

[0037] In one implementation scenario, the user information matched by the user is actively input by the user, and the user information includes at least one of the user's identity, occupation, and hobbies. For each target task corresponding to the user information, the scheduling module is used to analyze the user information, thereby determining the task identifier matched by the user information, and taking the candidate parameter set corresponding to the task identifier as the target parameter set. Then, according to the obtained target parameter set, the constructed target adjustment subnetwork is obtained.

[0038] In another embodiment, the user information of the user is determined by the intelligent terminal through analysis based on the historical input information of the user. For example, if the user asks questions related to medicine on the intelligent terminal many times, the intelligent terminal will analyze the specific content of the user's questions and determine that the user information of the user is a medical technician, a disease patient, or a medical student.

[0039] It should be noted that before obtaining the user's historical input information, the smart terminal sends an information acquisition request to the user in advance and informs the user of the specific purpose of the information. After the user grants the smart terminal information acquisition authority, the smart terminal obtains the user's historical input information based on the information acquisition authority.

[0040] S202: Input the task processing instruction into the intelligent analysis model, and use the model basic network and target adjustment sub-network in the intelligent analysis model to obtain the processing content matching each target task.

[0041] In one embodiment, the task processing instructions are input into the intelligent analysis model so that the model base network analyzes the task processing instructions in combination with the target adjustment subnetwork, and finally outputs the processing content matching each target task at the same time to improve the processing efficiency of the target tasks.

[0042] The above scheme builds a target adjustment sub-network with a higher matching degree with the corresponding user based on the user information, so that the intelligent analysis model including the model basic network and the target adjustment sub-network can process the target tasks of the corresponding user with higher accuracy, thus improving the user experience. In addition, since it is not necessary to analyze each target task to determine its corresponding type or field, the computing resource consumption of the intelligent analysis model is reduced.

[0043] See also Figure 3 , Figure 3 yes Figure 2 Step S201 corresponds to a flowchart of another implementation. Specifically, the model base network includes multiple analysis nodes, at least some of the analysis nodes are matched with at least one initial adjustment sub-network, and the implementation process of step S201 includes:

[0044] S301: In response to the initial adjustment sub-network matching having at least one candidate parameter set, a scheduling module is used to select a parameter set that matches the user information from all candidate parameter sets matched by the initial adjustment sub-network as a target parameter set.

[0045] In one embodiment, the model base network includes a plurality of analysis nodes, at least one analysis node is matched with a predetermined initial adjustment subnetwork, and each initial adjustment subnetwork is matched with at least one candidate parameter set obtained after training, and the candidate parameter set matches a preset task identifier. For all candidate parameter sets matched by each initial adjustment subnetwork, a scheduling module is used to select a candidate parameter set that matches each user information as a target parameter set.

[0046] In one implementation scenario, the model base network includes multiple network layers, each network layer is used as an analysis node, and at least some of the analysis nodes are arranged in parallel with the initial regulation subnetwork. Among them, the analysis node can be arranged in parallel with one or more initial regulation subnetworks. When the same analysis node matches multiple initial regulation subnetworks, the multiple initial regulation subnetworks matched by the analysis node are stacked. And, the matching relationship between the analysis node and the initial regulation subnetwork is determined based on the LoRA algorithm, that is, the position of each initial regulation subnetwork is determined according to the fine-tuning process of the model base network by the LoRA algorithm.

[0047] In a specific application scenario, when the model base network is a Transformer network, it includes an input part, multiple stacked encoders, multiple stacked decoders, and an output part, and each sublayer included in the input part, encoder, decoder, and output part is used as an analysis node. The structure of each sublayer in the Transformer network can refer to the existing model structure and will not be elaborated in detail here.

[0048] In another embodiment, at least one analysis node is coupled to a storage module, in which at least one candidate parameter set matching the corresponding analysis node is stored. The scheduling module selects a candidate parameter set matching each user information from the storage module matching the analysis node as the target parameter set.

[0049] In one implementation scenario, all analysis nodes are coupled to the same scheduling module, so that the scheduling module selects a candidate parameter set matching the user information from each storage module as the target parameter set. Alternatively, the analysis node is coupled with a corresponding storage module and a scheduling module, and the scheduling model is used to select a candidate parameter set matching the user information from the corresponding storage module as the target parameter set.

[0050] S302: Based on the selected target parameter set, construct a target adjustment subnetwork matching the corresponding analysis node.

[0051] In one embodiment, for each initial regulation subnetwork, the selected target parameter set is concatenated to obtain a regulation parameter matrix. The regulation parameter matrix is ​​added to the corresponding initial regulation subnetwork to obtain a constructed target regulation subnetwork. The regulation parameter matrix is ​​used to adjust the input of the corresponding target regulation subnetwork.

[0052] Specifically, each target parameter set contains multiple weight parameters, and all target parameter sets matched by each initial adjustment sub-network are spliced ​​to obtain an adjustment parameter matrix. The spliced ​​adjustment parameter matrix is ​​added to the corresponding initial adjustment sub-network to obtain the target adjustment sub-network, which has the ability to adjust the input content.

[0053] In one implementation scenario, the splicing order of the target parameter sets is consistent with the splicing order of the target tasks in the task processing instructions. And, the process of the target adjustment subnetwork adjusting the input content includes: multiplying the input of the target adjustment subnetwork with the corresponding adjustment parameter matrix.

[0054] In a specific application scenario, the user information corresponding to the task processing instructions is "medical technicians", "computer technicians" and "mechanical technicians" in order. The initial adjustment subnetwork matched by the "classification layer" analysis node in the model basic network includes multiple candidate parameter sets that match the task identifiers "medicine", "computer", "education" and "mechanical", etc. Then, the calling module is used to select the candidate parameter sets corresponding to the task identifiers "medicine", "computer" and "mechanical" as the target parameter sets, and the adjustment parameter matrix is ​​obtained after splicing according to the splicing order of the target tasks in the corresponding task processing instructions. The adjustment parameter matrix is ​​added to the initial adjustment subnetwork corresponding to the analysis node "classification layer" to obtain the target adjustment subnetwork.

[0055] In another embodiment, the specific implementation process of step S201 may further include: in response to each analysis node corresponding to a storage module, in which a corresponding candidate parameter set is stored, the scheduling module selects a target parameter set matching the user information from the storage module, and performs shielding processing on the candidate parameter set in the storage module that is different from the target parameter set. The shielding processing may be setting a corresponding shielding coefficient for the corresponding candidate parameter set, and the shielding coefficient is 0.

[0056] Furthermore, the scheduling module sorts the target parameter sets of the storage modules according to the splicing order of the target tasks in the corresponding task processing instructions, and uses the sorted storage modules as the target regulation subnetwork.

[0057] See also Figure 4 , Figure 4 yes Figure 2Step S202 in the flowchart corresponds to another embodiment. Specifically, the implementation process of step S202 includes:

[0058] S401: Obtain the historical output results of the previous analysis node output.

[0059] In one embodiment, in response to the model base network including a plurality of stacked analysis nodes, and the input of the current analysis node is the output of the previous analysis node, the historical output result of the previous analysis node output is obtained, wherein the previous analysis node is located before the current analysis node and is adjacent to the current analysis node.

[0060] S402: Input the historical output results to the current analysis node to obtain the current output results output by the current analysis node, until the processing content matching each target task is obtained.

[0061] In one embodiment, the historical output result output by the previous analysis node is input to the current analysis node, so that the current analysis node processes the historical output result and then outputs the current output result.

[0062] In one implementation scenario, in response to the current analysis node not matching the target regulation subnetwork, the historical output result is input to the current analysis node, and the current analysis node is directly used to process the input historical output result to obtain the current output result output by the current analysis node.

[0063] Alternatively, in response to the current analysis node matching a target regulation subnetwork, the historical output results are input to the current analysis node, and the output of the current analysis node is adjusted in combination with the target regulation subnetwork matching the current analysis node, and finally the current output result of the current analysis node is obtained.

[0064] Specifically, the historical output results are respectively input to the current analysis node and its matching target regulation sub-network, and a first reference result output by the current analysis node and a second reference result output by the target regulation sub-network matching the current analysis node are obtained. Based on the first reference result and the second reference result, the current output result output by the current analysis node is obtained. The current output result is obtained by fusing or splicing the first reference result and the second reference result.

[0065] In a specific implementation scenario, please refer to Figure 5 , Figure 5It is a schematic diagram of a specific implementation scenario corresponding to step S402. The current analysis node is matched with a stacked first target regulation subnetwork and a second target regulation subnetwork. After obtaining the historical output results output by the previous analysis node, the historical output results are respectively input into the current analysis node in the model basic network and the first target regulation subnetwork matched by the current analysis node. The current analysis node analyzes the historical output content to obtain a first reference result. The first target regulation subnetwork analyzes the historical output content, and inputs the obtained analysis content into the second target regulation subnetwork. The second target regulation subnetwork further analyzes the analysis content output by the first target regulation subnetwork, and finally outputs a second reference result. The first reference result and the second reference result are fused or spliced, and the result obtained after fusion or splicing is used as the current output result output by the current analysis node.

[0066] S403: Update the current analysis node to the previous analysis node, update the current output result to the historical output result, and return to obtain the historical output result output by the previous analysis node until the processing content matched by each target task is obtained.

[0067] In one embodiment, in response to the current analysis node not being the last analysis node in the model base network, the current analysis node is updated to the previous analysis node, and the current output result is updated to the historical output result, and the process returns to step S401 to execute subsequent steps in sequence until the processing content matching each target task is finally obtained.

[0068] Alternatively, in response to the current analysis node being the last analysis node in the model base network, the current analysis result output by the current analysis node is used as the processing content for matching each target task.

[0069] In another embodiment, the task processing instruction may also be determined based on at least one target task input by the same user. Based on this, the specific implementation process of step S102 includes: using the scheduling module to select a candidate parameter set matching each target task as the target parameter set, and constructing a target adjustment subnetwork according to the target parameter set.

[0070] Specifically, after obtaining at least one target task input by the same user, a task processing instruction is generated according to all target tasks. After the task processing instruction is input into the intelligent analysis model, the task processing instruction is analyzed by the scheduling module, and the corresponding task identifier is determined according to the semantic information of the target task, so as to select a target parameter set matching each target task from multiple candidate parameter sets according to the task identifier, and construct a target adjustment subnetwork according to the target parameter set. Among them, the specific process of constructing the target adjustment subnetwork can refer to the corresponding implementation method above, which will not be repeated here.

[0071] For example, when the target tasks input by the user include "What's the weather like today" and "Help me make a travel plan", the target tasks are analyzed to determine that their corresponding task identifiers are "weather query" and "text generation", and the scheduling module is used to select the corresponding target parameter set to construct the target regulation subnetwork.

[0072] Furthermore, the task processing instructions are input into the intelligent analysis model, and the model basic network and target adjustment subnetwork in the intelligent analysis model are used to obtain the processing content matching each target task. The specific implementation process can refer to the corresponding implementation method above, and will not be elaborated in detail here.

[0073] In another embodiment, when the task processing instruction is determined based on the target tasks input by different users, the corresponding task identifier can also be determined based on the semantic information of each target task, so as to select a target parameter set matching each target task from multiple candidate parameter sets based on the task identifier, and construct a target adjustment subnetwork based on the target parameter set.

[0074] See also Figure 6 , Figure 6 FIG. 1 is a schematic diagram of a structure of an implementation of a task processing system of the present application. The task processing system comprises a coupled acquisition module 10 and a processing module 20 .

[0075] Specifically, the acquisition module 10 is used to acquire a task processing instruction triggered by at least one target task.

[0076] The processing module 20 is used to input the task processing instructions into the intelligent analysis model to obtain the processing content output by the intelligent analysis model that matches each target task; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to the scheduling module, and the scheduling module is used to select a target parameter set that matches the target task from multiple trained candidate parameter sets to construct a target adjustment sub-network.

[0077] In one embodiment, the task processing instructions are determined based on target tasks input by multiple users, and each user is matched with corresponding user information. Based on this, the processing module 20 inputs the task processing instructions into the intelligent analysis model to obtain the processing content output by the intelligent analysis model that matches each target task, including: using the scheduling module to select a candidate parameter set that matches the user information as the target parameter set, and constructing a target adjustment subnetwork based on the target parameter set; inputting the task processing instructions into the intelligent analysis model, and using the model basic network and the target adjustment subnetwork in the intelligent analysis model to obtain the processing content that matches each target task.

[0078] In one embodiment, the model basic network includes multiple analysis nodes, at least some of the analysis nodes are matched with at least one initial adjustment sub-network, the processing module 20 uses the scheduling module to select a candidate parameter set that matches the user information as the target parameter set, and constructs a target adjustment sub-network according to the target parameter set, including: in response to the initial adjustment sub-network matching at least one candidate parameter set, using the scheduling module to select a parameter that matches the user information from all candidate parameter sets matched by the initial adjustment sub-network as the target parameter set; based on the selected target parameter set, construct a target adjustment sub-network that matches the corresponding analysis node.

[0079] In one embodiment, the processing module 20 constructs a target regulation subnetwork that matches the corresponding analysis node based on the selected target parameter set, including: splicing the selected target parameter set to obtain an adjustment parameter matrix; adding the adjustment parameter matrix to the corresponding initial regulation subnetwork to obtain a target regulation subnetwork; wherein the adjustment parameter matrix is ​​used to adjust the input of the corresponding target regulation subnetwork.

[0080] In one embodiment, the processing module 20 uses the model base network and the target adjustment subnetwork in the intelligent analysis model to obtain the processing content matching each target task, including: obtaining the historical output result of the previous analysis node output; in response to the current analysis node not matching the target adjustment subnetwork, inputting the historical output result to the current analysis node to obtain the current output result of the current analysis node output; in response to the current analysis node matching the target adjustment subnetwork, inputting the historical output result to the current analysis node, and adjusting the output of the current analysis node using the target adjustment subnetwork matching the current analysis node to obtain the current output result of the current analysis node output; updating the current analysis node to the previous analysis node, updating the current output result to the historical output result, and returning to obtaining the historical output result of the previous analysis node output, until the processing content matching each target task is obtained.

[0081] In one embodiment, the processing module 20 inputs the historical output results to the current analysis node, and uses the target regulation subnetwork matched by the current analysis node to regulate the output of the current analysis node to obtain the current output result of the current analysis node, including: inputting the historical output results to the current analysis node and its matching target regulation subnetwork respectively, obtaining a first reference result of the output of the current analysis node and a second reference result of the output of the target regulation subnetwork matched by the current analysis node; based on the first reference result and the second reference result, obtaining the current output result of the output of the current analysis node.

[0082] In one embodiment, the task processing instruction is determined based on at least one target task input by the same user. Based on this, the processing module 20 inputs the task processing instruction into the intelligent analysis model to obtain the processing content output by the intelligent analysis model that matches each target task, including: using the scheduling module to select a candidate parameter set that matches each target task as a target parameter set, and constructing a target adjustment subnetwork based on the target parameter set; inputting the task processing instruction into the intelligent analysis model, and using the model basic network and the target adjustment subnetwork in the intelligent analysis model to obtain the processing content that matches each target task.

[0083] See also Figure 7 , Figure 7 : It is a structural diagram of an embodiment of an electronic device of the present application. The electronic device includes: a memory 30 and a processor 40 coupled to each other. The memory 30 stores program instructions, and the processor 40 is used to execute the program instructions to implement the method mentioned in any of the above embodiments. Specifically, the electronic device includes but is not limited to: a desktop computer, a laptop computer, a tablet computer, a server, etc., which are not limited here. In addition, the processor 40 can also be called a CPU (Center Processing Unit). The processor 40 may be an integrated circuit chip with signal processing capabilities. The processor 40 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 40 can be implemented by an integrated circuit chip.

[0084] See also Figure 8 , Figure 8 It is a structural diagram of an embodiment of a computer-readable storage medium of the present application. The computer-readable storage medium 50 stores program instructions 60 that can be executed by a processor. When the program instructions 60 are executed by the processor, the method mentioned in any of the above embodiments is implemented.

[0085] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0086] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0087] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0088] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0089] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A task processing method, characterized in that: include: Obtaining a task processing instruction triggered by at least one target task; The task processing instruction is input into the intelligent analysis model to obtain the processing content matched with each target task output by the intelligent analysis model; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to a scheduling module, and the scheduling module is used to select a target parameter set matching the target task from a plurality of trained candidate parameter sets to construct the target adjustment sub-network.

2. The method according to claim 1, characterized in that The task processing instruction is determined based on target tasks input by multiple users, and each of the users is matched with corresponding user information, and the task processing instruction is input into the intelligent analysis model to obtain the processing content matched with each target task output by the intelligent analysis model, including: Using the scheduling module to select the candidate parameter set matching the user information as the target parameter set, and constructing the target adjustment subnetwork according to the target parameter set; The task processing instructions are input into the intelligent analysis model, and the model basic network and the target adjustment subnetwork in the intelligent analysis model are used to obtain the processing content matching each target task.

3. The method according to claim 2, characterized in that The model basic network includes a plurality of analysis nodes, at least some of the analysis nodes are matched with at least one initial adjustment sub-network, the scheduling module is used to select the candidate parameter set matching the user information as the target parameter set, and the target adjustment sub-network is constructed according to the target parameter set, including: In response to the initial adjustment subnetwork matching at least one of the candidate parameter sets, using the scheduling module to select a parameter set that matches the user information from all the candidate parameter sets matched by the initial adjustment subnetwork as the target parameter set; Based on the selected target parameter set, the target regulation subnetwork matching the corresponding analysis node is constructed.

4. The method according to claim 3, characterized in that The step of constructing the target adjustment subnetwork matching the corresponding analysis node based on the selected target parameter set includes: The selected target parameter sets are spliced ​​to obtain an adjustment parameter matrix; The adjustment parameter matrix is ​​added to the corresponding initial adjustment sub-network to obtain the target adjustment sub-network; wherein the adjustment parameter matrix is ​​used to adjust the input of the corresponding target adjustment sub-network.

5. The method according to claim 3, characterized in that: The using the model basic network and the target adjustment subnetwork in the intelligent analysis model to obtain processing content matching each target task includes: Get the historical output results of the previous analysis node; In response to the current analysis node not matching the target regulation sub-network, inputting the historical output result to the current analysis node to obtain a current output result output by the current analysis node; In response to the current analysis node being matched with the target adjustment sub-network, the historical output result is input to the current analysis node, and the output of the current analysis node is adjusted using the target adjustment sub-network matched by the current analysis node to obtain the current output result output by the current analysis node; The current analysis node is updated to the previous analysis node, the current output result is updated to the historical output result, and the process of obtaining the historical output result of the previous analysis node output is returned until the processing content matched by each target task is obtained.

6. The method according to claim 5, characterized in that The inputting of the historical output result to the current analysis node, and adjusting the output of the current analysis node using the target adjustment sub-network matched by the current analysis node to obtain the current output result of the current analysis node output, includes: Input the historical output results to the current analysis node and the target regulation sub-network matched thereto respectively, and obtain a first reference result output by the current analysis node and a second reference result output by the target regulation sub-network matched thereto; Based on the first reference result and the second reference result, a current output result output by the current analysis node is obtained.

7. The method according to claim 1, characterized in that The task processing instruction is determined based on at least one target task input by the same user, and the task processing instruction is input into the intelligent analysis model to obtain the processing content output by the intelligent analysis model that matches each target task, including: Using the scheduling module to select the candidate parameter set matching each of the target tasks as the target parameter set, and constructing the target adjustment subnetwork according to the target parameter set; The task processing instructions are input into the intelligent analysis model, and the model basic network and the target adjustment subnetwork in the intelligent analysis model are used to obtain the processing content matching each target task.

8. A task processing system, characterized in that: include: An acquisition module, used for acquiring a task processing instruction triggered by at least one target task; A processing module is used to input the task processing instructions into the intelligent analysis model to obtain the processing content output by the intelligent analysis model that matches each target task; wherein the intelligent analysis model includes a model base network and at least one target adjustment sub-network parallel to the model base network, the model base network is coupled to a scheduling module, and the scheduling module is used to select a target parameter set that matches the target task from multiple trained candidate parameter sets to construct the target adjustment sub-network.

9. An electronic device, characterized in that: include: A memory and a processor coupled to each other, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the task processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the task processing method according to any one of claims 1 to 7 is implemented.

Images