Execution method and device of intelligent model

Abstract

The invention discloses an execution method and device of an intelligent model, and the method comprises the steps: obtaining a target processing intention in response to a processing request sent by any execution terminal; according to the target processing intention and description files of a plurality of intelligent models, calling a target model to obtain a processing result output by the target model; wherein the target model is a model, matched with the target processing intention, in the multiple intelligent models; the description file of the intelligent model at least represents an execution terminal where the intelligent model is installed and functions which can be realized by the intelligent model; the execution terminal installed on the intelligent model is determined based on the equipment matching degree between the intelligent model and the execution terminal.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a method and device for executing an intelligent model. Background Art

[0002] With the rapid development of artificial intelligence technology, more and more small artificial intelligence models can be directly run on user devices, such as smartphones, tablets, and wearable devices. These models provide users with various intelligent services such as speech recognition, image recognition, and natural language processing by processing data in real time, significantly improving the work efficiency and life convenience of users.

[0003] However, the installation and maintenance of these models rely on manual operations by users, and the process is complex and error-prone. Summary of the Invention

[0004] In view of this, this application provides a method and device for executing an intelligent model as follows:

[0005] A method for executing an intelligent model includes:

[0006] Obtaining a target processing intention in response to a processing request sent by any execution terminal;

[0007] Invoking a target model according to the target processing intention and description files of multiple intelligent models to obtain a processing result output by the target model;

[0008] Wherein, the target model is a model that matches the target processing intention among the multiple intelligent models; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

[0009] In the above method, preferably, the device matching degree between the intelligent model and the execution terminal is obtained by the following method:

[0010] Obtaining the component matching degrees of the intelligent model and the execution terminal in multiple matching dimensions;

[0011] Processing the component matching degrees according to the dimension weight values corresponding to the matching dimensions to obtain the device matching degree between the intelligent model and the execution terminal.

[0012] In the above method, preferably, the dimension weight values corresponding to the matching dimensions correspond to the functions that the intelligent model can achieve.

[0013] Preferably, the above method for obtaining the component matching degrees of the intelligent model and the execution terminal in multiple matching dimensions includes:

[0014] Comparing the requirement parameters of at least one operation requirement item of the intelligent model in the matching dimension with the configuration parameters of the corresponding terminal configuration item of the operation requirement item on the execution terminal to obtain a comparison result; the comparison result represents the degree of difference between the requirement parameters and the configuration parameters;

[0015] Adjusting the initial matching degree in the matching dimension according to the comparison result to obtain the component matching degree of the intelligent model and the execution terminal in the matching dimension.

[0016] Preferably, the above method further includes:

[0017] When the execution terminal has a user authorization identifier, reading the configuration parameters of the terminal configuration item on the execution terminal;

[0018] Wherein, the user authorization identifier indicates that the user to whom the execution terminal belongs at least allows the configuration parameters to be read.

[0019] Preferably, adjusting the initial matching degree in the matching dimension according to the comparison result includes:

[0020] When the comparison result represents that the requirement parameters are superior to the configuration parameters, reducing the corresponding value of the initial matching degree in the matching dimension according to the degree of difference represented by the comparison result;

[0021] When the comparison result represents that the requirement parameters are inferior to the configuration parameters, increasing the corresponding value of the initial matching degree in the matching dimension according to the degree of difference represented by the comparison result;

[0022] When the comparison result represents that the requirement parameters are consistent with the configuration parameters, maintaining the initial matching degree in the matching dimension.

[0023] Preferably, the intelligent model is determined by the following method:

[0024] Obtaining the meeting type of the meeting to which the execution terminal belongs;

[0025] According to the meeting type, screening candidate models that match the meeting type in a model library containing multiple candidate models;

[0026] Processing the candidate model to match the execution terminal to obtain the intelligent model.

[0027] In the above method, preferably, the intelligent model matches the meeting type, including: the intelligent model can execute any meeting task involved in the meeting type;

[0028] Among them, processing the candidate model to match the execution terminal to obtain the intelligent model includes:

[0029] Adjusting the model parameters of the candidate model so that the data processing accuracy of the candidate model matches the execution ability of the execution terminal.

[0030] In the above method, preferably, according to the target processing intention and the description files of multiple intelligent models, calling the target model to obtain the processing result output by the target model includes:

[0031] Providing the target processing intention and the description files of multiple intelligent models to the inference model, and the inference model calls the target model that matches the target processing intention according to the description files through an execution function, so that the target model outputs a processing result.

[0032] An execution device for an intelligent model includes:

[0033] An intention acquisition unit, configured to obtain a target processing intention in response to a processing request sent by any execution terminal;

[0034] A model calling unit, configured to call a target model according to the target processing intention and the description files of multiple intelligent models to obtain the processing result output by the target model;

[0035] Among them, the target model is a model in the multiple intelligent models that matches the target processing intention; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can implement; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

[0036] A computer device / system, including: a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the execution method of the intelligent model described in any one of the above.

[0037] A computer-readable storage medium, on which a computer program / instruction is stored, and when the computer program / instruction is executed by a processor, the execution method of the intelligent model described in any one of the above is implemented.

[0038] A computer program product, including a computer program / instruction, and when the computer program / instruction is executed by a processor, the execution method of the intelligent model described in any one of the above is implemented.

[0039] As can be seen from the above technical solution, in an execution method and device for an intelligent model disclosed in this application, the intelligent model is installed on the corresponding execution terminal based on the device matching degree with the execution terminal, and the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve are represented by a description file. Thus, when any execution terminal sends a processing request, the intelligent model matching the target processing intention can be called according to the corresponding target processing intention and the description files of each intelligent model, so as to obtain the processing results output by these intelligent models. It can be seen that in this application, there is no need for the user to manually install the intelligent model or manually call the intelligent model, and the processing result can be obtained. Therefore, while reducing the complexity of user operations, the accuracy of processing can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0041] Figure 1 It is a flowchart of an execution method for an intelligent model provided by an embodiment of this application;

[0042] Figure 2 It is an example diagram of an execution terminal in an embodiment of this application;

[0043] Figure 3 It is a partial flowchart of an execution method for an intelligent model provided by an embodiment of this application;

[0044] Figure 4 It is another partial flowchart of an execution method for an intelligent model provided by an embodiment of this application;

[0045] Figure 5 It is yet another partial flowchart of an execution method for an intelligent model provided by an embodiment of this application;

[0046] Figure 6 It is a schematic structural diagram of an execution device for an intelligent model provided by an embodiment of this application;

[0047] Figure 7 It is another schematic structural diagram of an execution device for an intelligent model provided by an embodiment of this application;

[0048] Figure 8 It is a schematic structural diagram of an electronic device provided by an embodiment of this application;

[0049] Figure 9This application is applicable to the flowchart of model deployment and invocation in the scenario of holding product design meetings. Detailed implementation manners

[0050] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0051] Refer to Figure 1 As shown, it is the implementation flowchart of an execution method of an intelligent model provided by an embodiment of the present application. This method can be applicable to an electronic device capable of data processing, such as any execution terminal installed with an intelligent model, or a server connected to the execution terminal, such as Figure 2 shown in. The technical solution in this embodiment is mainly used to reduce the operation complexity of users in processing intelligent models.

[0052] Specifically, the method in this embodiment may include the following steps:

[0053] Step 101: In response to a processing request sent by any execution terminal, obtain a target processing intention.

[0054] Specifically, in this embodiment, an intention recognition model may be used to process the processing request to obtain a target processing intention, such as the intention of generating text, the intention of generating an image, the intention of speech recognition, etc.

[0055] For example, as Figure 2 shown in, there are three execution terminals, namely a mobile phone, a tablet device, and a notebook, accessing the conference server. The mobile phone sends a processing request, and the conference server obtains the target processing intention of generating an image according to the processing request.

[0056] Step 102: According to the target processing intention and the description files of multiple intelligent models, call the target model to obtain the processing result output by the target model.

[0057] Among them, the target model is the model that matches the target processing intention among multiple intelligent models. The description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can implement. And the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

[0058] For example, as Figure 2As shown, in a mobile phone, a tablet device, and a laptop, the installed intelligent models are determined respectively based on the device matching degrees between them and respective intelligent models, and the installed intelligent models in the mobile phone, the tablet device, and the laptop each have a description file. The description file can be a description file in JSON format. For example, the description file of the intelligent model installed on the mobile phone represents that the intelligent model is installed on the mobile phone, and represents that the function implemented by the intelligent model installed on the mobile phone is the function of text generation; the description file of the intelligent model installed on the tablet represents that the intelligent model is installed on the tablet device, and represents that the function implemented by the intelligent model installed on the tablet device is the function of speech recognition; the description file of the intelligent model installed on the laptop represents that the intelligent model is installed on the laptop, and represents that the function implemented by the intelligent model installed on the laptop is the function of image generation.

[0059] Based on this, after the mobile phone sends a processing request for image generation, the conference server obtains the target processing intention for generating an image according to the processing request. Based on the description files of the intelligent models installed on the mobile phone, the tablet device, and the laptop respectively, it can be determined that the intelligent model matching the target processing intention is the intelligent model for image generation, and this intelligent model is installed on the laptop. At this time, the intelligent model installed on the laptop can be called, and the intelligent model installed on the laptop generates an image based on the processing request to output the generated image.

[0060] Another example is that after the tablet device sends a processing request for text generation, the conference server obtains the target processing intention for generating text according to the processing request. Based on the description files of the intelligent models installed on the mobile phone, the tablet device, and the laptop respectively, it can be determined that the intelligent model matching the target processing intention is the intelligent model for text generation, and this intelligent model is installed on the mobile phone. At this time, the intelligent model installed on the mobile phone can be called, and the intelligent model installed on the mobile phone generates text based on the processing request to output the generated text.

[0061] In one implementation, in step 102, the target processing intention and the description files of multiple intelligent models can be provided to an inference model, and the inference model calls a target model that matches the target processing intention according to the description files by executing a function, so that the target model outputs a processing result.

[0062] Among them, the inference model can be a decision-making model based on a machine learning algorithm. The decision-making model can process the description files and the target processing intention through executing a function Function Call, so that the decision-making model can call a target model that matches the target processing intention, and thus the target model outputs a corresponding processing result.

[0063] As can be seen from the above technical solution, in the execution method of an intelligent model provided by the embodiments of the present application, the intelligent model is installed on the corresponding execution terminal based on the device matching degree with the execution terminal, and the execution terminal on which the intelligent model is installed and the functions that the intelligent model can implement are characterized by a description file. Thus, when any execution terminal sends a processing request, the intelligent model matching the target processing intention can be called according to the corresponding target processing intention and the description files of each intelligent model, so as to obtain the processing results output by these intelligent models. It can be seen that in this embodiment, the user does not need to manually install the intelligent model, nor does the user need to manually call the intelligent model, and the processing result can be obtained. Thus, while reducing the complexity of user operations, the accuracy of processing can be improved.

[0064] In one implementation, the device matching degree between the intelligent model and the execution terminal can be obtained in the following manner, as Figure 3 shown in:

[0065] Step 301: Obtain the component matching degrees of the intelligent model and the execution terminal on multiple matching dimensions.

[0066] Among them, the matching dimensions can include at least multiple of hardware, software, performance, and tasks. The matching dimension of hardware refers to the dimension of hardware configurations such as processors, memory, and graphics cards. The matching dimension of software refers to the dimension of software configurations such as operating systems and installed software. The matching dimension of performance refers to the dimension of performance aspects such as processing speed and power consumption. The matching dimension of tasks can include dimensions of task aspects such as task type, required resources, and priority. The task type can include: task types of compute-intensive or I / O-intensive. The required resources of the task can include: resources such as processors and memory space. The priority of the task can include high, medium, and low priorities.

[0067] Specifically, in this embodiment, the component matching degrees of each intelligent model and each execution terminal on multiple matching dimensions can be obtained respectively.

[0068] For example, obtain the component matching degrees of the intelligent model for generating images with mobile phones, tablet devices, and laptops on the matching dimensions of hardware, software, performance, and tasks respectively; and obtain the component matching degrees of the intelligent model for generating text with mobile phones, tablet devices, and laptops on the matching dimensions of hardware, software, performance, and tasks respectively; and obtain the component matching degrees of the intelligent model for recognizing speech with mobile phones, tablet devices, and laptops on the matching dimensions of hardware, software, performance, and tasks respectively.

[0069] Step 302: Process the component matching degrees according to the dimension weight values corresponding to the matching dimensions to obtain the device matching degree between the intelligent model and the execution terminal.

[0070] It should be noted that the dimension weight values corresponding to the matching dimensions correspond to the functions that the intelligent model can achieve. For example, for an intelligent model that generates images, the dimension weight value corresponding to the hardware matching dimension is relatively high; for an intelligent model that generates text, the dimension weight value corresponding to the software matching dimension is relatively high; for an intelligent model that recognizes speech, the dimension weight value corresponding to the performance matching dimension is relatively high.

[0071] Specifically, in this embodiment, the dimension weight values corresponding to each matching dimension can be used to perform weighted summation on the component matching degrees on each matching dimension to obtain the device matching degree between the intelligent model and the intelligent terminal.

[0072] For example, taking the three matching dimensions of hardware, software, and performance as an example, the device matching degree is obtained through the following formula (1):

[0073] Device matching degree = (component matching degree on hardware × dimension weight value corresponding to hardware) + (component matching degree on software × dimension weight value corresponding to software) + (component matching degree on performance × dimension weight value corresponding to performance).

[0074] For example, through formula (1), in this embodiment, the component matching degrees of the intelligent model for generating images and the mobile phone on the three matching dimensions of hardware, software, and performance are weighted and summed according to the dimension weight values corresponding to these three matching dimensions, such as 0.6, 0.2, and 0.2, to obtain the device matching degree between the intelligent model for generating images and the mobile phone;

[0075] For example, through formula (1), in this embodiment, the component matching degrees of the intelligent model for generating images and the notebook on the three matching dimensions of hardware, software, and performance are weighted and summed according to the dimension weight values corresponding to these three matching dimensions, such as 0.6, 0.2, and 0.2, to obtain the device matching degree between the intelligent model for generating images and the notebook;

[0076] Again, for example, through formula (1), in this embodiment, the component matching degrees of the intelligent model for recognizing speech and the tablet device on the three matching dimensions of hardware, software, and performance are weighted and summed according to the dimension weight values corresponding to these three matching dimensions, such as 0.2, 0.3, and 0.5, to obtain the device matching degree between the intelligent model for recognizing speech and the tablet device;

[0077] Again, for example, through formula (1), in this embodiment, the component matching degrees of the intelligent model for recognizing speech and the notebook on the three matching dimensions of hardware, software, and performance are weighted and summed according to the dimension weight values corresponding to these three matching dimensions, such as 0.2, 0.3, and 0.5, to obtain the device matching degree between the intelligent model for recognizing speech and the notebook.

[0078] It can be seen that the device matching degree obtained in this embodiment represents the matching degree between the intelligent model and the execution terminal. The higher this matching degree is, the better the effect of implementing the corresponding function after the intelligent model is installed on the execution terminal. Therefore, in this embodiment, according to the device matching degree, the execution terminals where each intelligent model is installed are determined, and after receiving a processing request, a suitable target model is called from the intelligent models installed on these execution terminals according to the corresponding target processing intention to achieve the corresponding processing target.

[0079] Based on the above implementation, in step 301, the component matching degrees of the intelligent model and the execution terminal in multiple matching dimensions can be obtained through the following methods, such as Figure 4 shown in

[0080] Step 401: Compare the demand parameters of at least one running requirement item of the intelligent model in the matching dimension with the configuration parameters of the corresponding terminal configuration item of the running requirement item on the execution terminal to obtain a comparison result.

[0081] Among them, the comparison result represents the difference degree between the demand parameter and the configuration parameter.

[0082] For example, the running requirement items of the intelligent model on the hardware can include: demand items such as the processor, memory, and graphics card required by the intelligent model on the hardware. The corresponding terminal configuration items of the running requirement item on the execution terminal can include: terminal configuration items such as the processor, memory, and graphics card. Another example is that the running requirement items of the intelligent model on the software can include: demand items such as the operating system and installed software required by the intelligent model on the software. The corresponding terminal configuration items of the running requirement item on the execution terminal can include the operating system, installed software, etc. The running requirement items of the intelligent model in terms of performance can include: demand items such as the processing speed and power consumption required by the intelligent model in terms of performance. The corresponding terminal configuration items of the running requirement item on the execution terminal can include the processing speed, power consumption, etc.

[0083] In specific implementation, in step 401, the demand parameters of each running requirement item of the intelligent model in each matching dimension can be compared with the configuration parameters of its corresponding terminal configuration item to obtain the comparison result corresponding to each running requirement item in each matching dimension.

[0084] Step 402: Adjust the initial matching degree in the matching dimension according to the comparison result to obtain the component matching degree of the intelligent model and the execution terminal in the matching dimension.

[0085] Specifically, each matching dimension has an initial matching degree, which is a preset value. After obtaining the comparison result, the initial matching degree on the corresponding matching dimension is adjusted (such as increased or decreased) according to the difference degree between the requirement parameter and the configuration parameter in the comparison result, so as to obtain the component matching degree between the intelligent model and the execution terminal on the matching dimension.

[0086] For example, in the case where the comparison result indicates that the requirement parameter is superior to the configuration parameter, according to the difference degree represented by the comparison result, the initial matching degree on the matching dimension is decreased by a corresponding value, and the decreased value matches the difference amount between the configuration parameter and the requirement parameter; in the case where the comparison result indicates that the requirement parameter is inferior to the configuration parameter, according to the difference degree represented by the comparison result, the initial matching degree on the matching dimension is increased by a corresponding value, and the increased value matches the difference amount between the configuration parameter and the requirement parameter; in the case where the comparison result indicates that the requirement parameter is consistent with the configuration parameter, the initial matching degree on the matching dimension is maintained as the component matching degree.

[0087] Based on the above implementation, in order to obtain the configuration parameter of the terminal configuration item corresponding to the operation requirement item of the execution terminal, user authorization is required. Specifically, in this embodiment, when the execution terminal has a user authorization identifier, the configuration parameter of the terminal configuration item is read on the execution terminal.

[0088] Among them, the user authorization identifier indicates that the user to whom the execution terminal belongs at least allows the configuration parameter to be read. Further, the user authorization identifier also indicates that the execution terminal is allowed to install the intelligent model. The user authorization identifier also indicates that the installed intelligent model is allowed to be called.

[0089] It should be noted that the user authorization identifier can be set by the user to whom the execution terminal belongs when the execution terminal accesses the conference server.

[0090] In one implementation manner, the intelligent model in this embodiment can be determined in the following manner, as Figure 5 shown in

[0091] Step 501: Obtain the conference type of the conference to which the execution terminal belongs.

[0092] Among them, the conference type can be understood as the type of the conference scenario, such as a seminar on the game development interface or a market survey conference on cosmetics, etc. The conference type determines the intelligent model to be used.

[0093] Step 502: According to the conference type, screen the candidate models that match the conference type in the model library containing multiple candidate models.

[0094] Among them, multiple candidate models are deployed in the model library, and each candidate model is an intelligent model capable of implementing corresponding functions. In this embodiment, candidate models whose implemented function types can match the meeting type are screened from the model library according to the meeting type, that is, candidate models.

[0095] For example, based on the meeting type of a seminar meeting on a game development interface, candidate models that match this meeting type are screened from the model library, such as intelligent models for generating images and intelligent models for recognizing speech.

[0096] Step 503: Process the candidate model to match the execution terminal to obtain an intelligent model.

[0097] It should be noted that in order to enable the execution terminal to better run the candidate model, the candidate model can be processed to match the execution terminal to obtain an intelligent model.

[0098] Among them, the intelligent model matching the meeting type can be: the intelligent model can execute the meeting tasks involved in the meeting type. Based on this, in step 503, the model parameters of the candidate model can be adjusted so that the data processing accuracy of the candidate model matches the execution ability of the execution terminal. Thus, the execution terminal can run the candidate model with adjusted data processing accuracy (higher or lower) with the current terminal configuration items.

[0099] For example, in this embodiment, the model parameters of the intelligent model for generating images can be adjusted so that the data processing accuracy of the obtained intelligent model is reduced, thereby reducing the data processing volume of the intelligent model and also reducing the performance consumption of the execution terminal, so as to be able to run faster in the seminar meeting on the game development interface and quickly provide the generated images to the participants.

[0100] Reference Figure 6 , is a schematic structural diagram of an execution device for an intelligent model provided by an embodiment of the present application. This device can be configured in an electronic device capable of data processing, such as any execution terminal installed with an intelligent model, or a server connected to the execution terminal, as Figure 2 shown in. The technical solution in this embodiment is mainly used to reduce the operation complexity of users in processing intelligent models.

[0101] Specifically, the device in this embodiment may include the following units:

[0102] An intention acquisition unit 601, configured to acquire a target processing intention in response to a processing request sent by any execution terminal;

[0103] A model invocation unit 602, configured to invoke a target model according to the target processing intention and description files of multiple intelligent models to obtain a processing result output by the target model;

[0104] Among them, the target model is the model among the multiple intelligent models that matches the target processing intention; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

[0105] As can be seen from the above technical solution, in an execution device of an intelligent model provided by an embodiment of the present application, by installing the intelligent model on the corresponding execution terminal based on the device matching degree between the intelligent model and the execution terminal, and characterizing the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve through the description file, when any execution terminal sends a processing request, the intelligent model that matches the target processing intention can be called according to the corresponding target processing intention and the description files of each intelligent model, so as to obtain the processing results output by these intelligent models. It can be seen that in this embodiment, there is no need for the user to manually install the intelligent model or manually call the intelligent model, and the processing result can be obtained, thereby reducing the user operation complexity and improving the processing accuracy.

[0106] In one implementation manner, the device in this embodiment may further include the following units, as Figure 7 shown in

[0107] A matching degree obtaining unit 603, configured to obtain the device matching degree between the intelligent model and the execution terminal in the following manner: obtain the component matching degrees of the intelligent model and the execution terminal on multiple matching dimensions; process the component matching degrees according to the dimension weight values corresponding to the matching dimensions to obtain the device matching degree between the intelligent model and the execution terminal.

[0108] Among them, the dimension weight value corresponding to the matching dimension corresponds to the functions that the intelligent model can achieve.

[0109] In one implementation manner, when the matching degree obtaining unit 603 obtains the component matching degrees of the intelligent model and the execution terminal on multiple matching dimensions, it is specifically configured to: compare the demand parameters of at least one operation requirement item of the intelligent model on the matching dimension with the configuration parameters of the terminal configuration item corresponding to the operation requirement item on the execution terminal to obtain a comparison result; the comparison result characterizes the difference degree between the demand parameters and the configuration parameters; adjust the initial matching degree on the matching dimension according to the comparison result to obtain the component matching degree of the intelligent model and the execution terminal on the matching dimension.

[0110] In one implementation, the matching degree obtaining unit 603 is further configured to: when the execution terminal has a user authorization identifier, read the configuration parameters of the terminal configuration item on the execution terminal; wherein, the user authorization identifier indicates that the user to whom the execution terminal belongs at least allows the configuration parameters to be read.

[0111] In one implementation, when the matching degree obtaining unit 603 adjusts the initial matching degree on the matching dimension according to the comparison result, it is specifically configured to: when the comparison result indicates that the requirement parameter is superior to the configuration parameter, reduce the corresponding value of the initial matching degree on the matching dimension according to the difference degree indicated by the comparison result; when the comparison result indicates that the requirement parameter is inferior to the configuration parameter, increase the corresponding value of the initial matching degree on the matching dimension according to the difference degree indicated by the comparison result; when the comparison result indicates that the requirement parameter is consistent with the configuration parameter, keep the initial matching degree on the matching dimension.

[0112] In one implementation, the model calling unit 602 is further configured to determine the intelligent model in the following manner: obtain the meeting type of the meeting to which the execution terminal belongs; according to the meeting type, screen candidate models that match the meeting type in a model library containing multiple candidate models; process the candidate models to match the execution terminal to obtain the intelligent model.

[0113] In one implementation, the intelligent model matching the meeting type includes: the intelligent model can execute any meeting tasks involved in the meeting type;

[0114] Wherein, when the model calling unit 602 processes the candidate model to match the execution terminal to obtain the intelligent model, it is specifically configured to: adjust the model parameters of the candidate model so that the data processing accuracy of the candidate model matches the execution ability of the execution terminal.

[0115] In one implementation, the model calling unit 602 is specifically configured to: provide the target processing intention and the description files of multiple intelligent models to the inference model, and the inference model calls the target model that matches the target processing intention according to the description files through an execution function so that the target model outputs a processing result.

[0116] It should be noted that the specific implementation manners of the units in this embodiment can refer to the corresponding contents in the foregoing, and will not be elaborated herein.

[0117] Reference Figure 8 , is a schematic structural diagram of an electronic device provided by an embodiment of the present application. The electronic device can be Figure 2For the conference server shown in , the electronic device in this embodiment may include the following structure:

[0118] A memory 801 for storing computer programs and data generated by the running of the computer programs;

[0119] A processor 802 for executing the computer program to achieve: in response to a processing request sent by any execution terminal, obtaining a target processing intention; according to the target processing intention and the description files of multiple intelligent models, calling a target model to obtain a processing result output by the target model; wherein, the target model is a model in the multiple intelligent models that matches the target processing intention; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

[0120] As can be seen from the above technical solution, in an electronic device provided in an embodiment of the present application, by installing an intelligent model on a corresponding execution terminal based on the device matching degree between the intelligent model and the execution terminal, and characterizing the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve through a description file, when any execution terminal sends a processing request, a target intelligent model that matches the target processing intention can be called according to the corresponding target processing intention and the description files of each intelligent model, so as to obtain the processing results output by these intelligent models. It can be seen that in this embodiment, there is no need for the user to manually install the intelligent model or manually call the intelligent model, and the processing result can be obtained, thereby reducing the user operation complexity and improving the processing accuracy.

[0121] Taking the scenario of a certain company holding a new product design meeting as an example, the technical solution of the present application will be described below:

[0122] A certain company is about to hold a new product design meeting in Conference Room A. In order to improve the efficiency and quality of the meeting, the organizer has obtained in advance the information of the participants and the devices they carry, including high-performance laptops, smartphones, etc. In order to make full use of the performance of these devices and ensure data security, the organizer plans to deploy a series of edge models (i.e., the intelligent models mentioned above) to locally run tasks of multiple intelligent models such as meeting minutes, document rewriting, translation, presentation program generation, architecture diagram generation, and PPT beautification on the devices.

[0123] However, the edge device (i.e., the execution terminal mentioned above) needs to automatically install and deploy a suitable local small model (i.e., the intelligent model mentioned above) to process the intelligent tasks required in the meeting and ensure that the meeting content is not exposed to the public network environment. Therefore, the company uses the collaborative scheduling method of multiple devices (i.e., multiple execution terminals) and multiple models described in this application. By intelligently selecting and scheduling the edge models, efficient resource allocation and task execution are achieved on multiple devices, while ensuring that data is processed locally to avoid the risk of leakage, thus meeting the needs of enterprises and individuals for intelligent applications.

[0124] Specifically, this application proposes an edge model selection and scheduling method based on local operation of multiple devices. The core of this method lies in intelligently identifying, selecting, and scheduling the edge models on multiple devices to achieve efficient resource allocation and task execution. The specific key points include:

[0125] 1. Intelligent identification and matching: The system can automatically identify information such as the computing power, storage space, operating system, and hardware configuration of each device, and intelligently match the most suitable edge model and device combination according to the task requirements.

[0126] 2. Innovation in intention understanding: By combining semantic analysis and interactive prompts, the complex intentions of users can be understood more accurately, and further confirmed through user feedback when necessary to ensure the accurate identification of task types.

[0127] 3. Dynamic scheduling and optimization: According to the priority, complexity, and resource requirements of tasks, the system dynamically schedules the operation of edge models on multiple devices to achieve the maximum utilization of resources and the efficient execution of tasks.

[0128] 4. Local data processing: All tasks and data are processed on local devices without relying on the cloud, ensuring data security and privacy protection and avoiding the risk of data leakage.

[0129] The advantages that can be obtained by adopting the technical solution of this application are as follows:

[0130] 1. Improve data security and privacy protection: Since all tasks and data are processed on local devices without relying on the cloud, the risk of data leakage is greatly reduced, enhancing data security and privacy protection.

[0131] 2. Improve task execution efficiency: By intelligently identifying and matching edge models and device combinations, and dynamically scheduling and optimizing resource allocation, the execution efficiency of tasks can be significantly improved, shortening the task completion time.

[0132] 3. Optimize resource utilization: Make full use of the computing power and resources of multiple devices to achieve the maximum utilization of resources and avoid resource waste and performance bottlenecks.

[0133] 4. Improve the accuracy of intent understanding: Through semantic analysis and interactive prompts, it is possible to understand user requests more deeply, reduce misunderstandings and misoperations, and enhance the user experience.

[0134] The following is a detailed description of the technical solution of this application:

[0135] I. Working principle of the solution:

[0136] The technical solution of this application aims to achieve end-side model selection and scheduling for multi-device local operation through a series of steps to meet the different needs of users in complex business scenarios. The core of the solution lies in intent understanding combining semantic analysis and interactive prompts, model quantization, intelligent agent JSON description, and task-oriented intelligent agent invocation. The following takes the scenario of a cross-regional product design meeting as an example to describe its working steps, as Figure 9 shown in:

[0137] 1. Determination of meeting requirements:

[0138] (1) Dialogue inquiry to collect meeting information: Collect meeting topics, goals, device information of participants, etc.

[0139] (2) Information collation: Collate it into a meeting requirements document.

[0140] 2. Model selection and quantization:

[0141] (1) Query the model library: Query the model library according to the requirements document and preliminarily screen candidate models.

[0142] (2) Model determination and quantization evaluation: Determine the required model and perform quantization processing on models that cannot run on the current hardware to facilitate the generation of a model list for JSON description.

[0143] 3. Device identification and generation of intelligent agent JSON description files:

[0144] (1) Device identification: Identify the available devices in the current environment and check the device status to generate a device description JSON.

[0145] (2) Model-device matching: Calculate the device matching degree of each device and the model according to formula (1).

[0146] (3) Combine with the external knowledge base to generate an intelligent agent JSON description file: Generate a new intelligent agent JSON description file for the matched model and device, including intelligent agent name, model information, device information, usage, resource consumption, etc.

[0147] 4. Intent understanding and task determination:

[0148] (1) User request reception: During a meeting, receive the user's request through a dialogue interface or a voice assistant.

[0149] (2) Intent understanding (semantic analysis): Use a local model to perform semantic analysis on the user's request to extract key information and potential intents, and determine what type of task the user wants to perform (such as document expansion, image generation, etc.). If the user's intent cannot be accurately determined, the system triggers an interactive prompt to request more information from the user.

[0150] 5. Agent selection and invocation:

[0151] (1) Agent selection: According to the result of intent understanding and the agent JSON description file, use the Function Call method to select an agent suitable for the current task.

[0152] 6. Result processing and display:

[0153] (1) Result collection and integration: After the agent processes the task, collect the results returned by the agent and perform integration and processing.

[0154] (2) Result display: Display the processed results to the user in an intuitive way.

[0155] II. Scheme implementation process:

[0156] First, the matching algorithm is described as follows:

[0157] The goal of this stage is to design and implement an algorithm that can accurately calculate the matching degree between a device and a task. This algorithm will consider the characteristics of the device, the requirements of the task, and relevant information in the external knowledge base to ensure the best pairing between the device and the task.

[0158] 1. Definition of characteristics and requirements:

[0159] First, this application needs to clarify the characteristics and requirements of the device and the task. For the device, this may include hardware configuration (such as Central Processing Unit (CPU), memory, graphics card, etc.), software environment (such as operating system, installed software, etc.), and performance parameters (such as processing speed, power consumption, etc.). For the task, this may include task type (such as computationally intensive, I / O intensive, etc.), required resources (such as CPU time, memory space, etc.), and priority (such as high, medium, low, etc.).

[0160] 2. Design of the matching formula:

[0161] To calculate the matching degree between a device and a task, this application can design a matching formula, such as formula (1). This formula will consider multiple factors and assign a weight to each factor to reflect its importance in the matching process.

[0162] Among them, the hardware matching degree, software matching degree, and performance matching degree (i.e., the component matching degree in the previous text) respectively represent the matching degree of the device with the task in terms of hardware, software, and performance. These matching degrees can be calculated by comparing the characteristics and requirements of the device and the task. The weight (i.e., the dimension weight value in the previous text) represents the relative importance of each factor in the matching process and can be adjusted according to the actual situation.

[0163] 3. Example

[0164] Taking a device with a high-performance independent graphics card as an example, assume there is an image generation task to be executed. This task has high requirements for the performance of the graphics card because the graphics card is responsible for processing image data.

[0165] (1) Hardware matching degree: Since the device has a high-performance independent graphics card, it is highly matched with the image generation task in terms of hardware. Based on this, a relatively high hardware matching degree value can be assigned to the graphics card performance.

[0166] (2) Software matching degree: If software related to the image generation task (such as an image processing library, rendering engine, etc.) is installed on the device, it also has a relatively high matching degree in terms of software. Otherwise, the software matching degree value needs to be reduced.

[0167] (3) Performance matching degree: In addition to the performance of the graphics card, the overall performance of the device, such as CPU speed, memory size, etc., also needs to be considered. If the overall performance of the device can meet the requirements of the image generation task, the performance matching degree is relatively high.

[0168] (4) Weight assignment: In this example, since the image generation task has high requirements for the performance of the graphics card, the hardware weight can be assigned a higher value. At the same time, since the task may also require certain software support and overall performance guarantee, the software and performance weights should not be too low.

[0169] It can be seen that by calculating the matching degree, the matching degree between this device and the image generation task can be obtained. If the matching degree is relatively high, this task can be assigned to this device for execution; if the matching degree is relatively low, it may be necessary to find other more suitable devices.

[0170] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0171] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0172] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0173] The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application will not be limited to the embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for executing an intelligent model, comprising: Obtaining a target processing intention in response to a processing request sent by any execution terminal; Invoking a target model according to the target processing intention and description files of multiple intelligent models to obtain a processing result output by the target model; Wherein, the target model is a model among the multiple intelligent models that matches the target processing intention; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can achieve; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

2. The method according to claim 1, wherein the device matching degree between the intelligent model and the execution terminal is obtained by the following method: Obtaining the component matching degrees of the intelligent model and the execution terminal on multiple matching dimensions; Processing the component matching degrees according to the dimension weight values corresponding to the matching dimensions to obtain the device matching degree between the intelligent model and the execution terminal.

3. The method according to claim 2, wherein the dimension weight value corresponding to the matching dimension corresponds to the function that the intelligent model can achieve.

4. The method according to claim 2, obtaining the component matching degrees of the intelligent model and the execution terminal on multiple matching dimensions, comprising: Comparing the demand parameters of at least one operation requirement item of the intelligent model on the matching dimension with the configuration parameters of the terminal configuration item corresponding to the operation requirement item on the execution terminal to obtain a comparison result; the comparison result characterizes the difference degree between the demand parameters and the configuration parameters; Adjusting the initial matching degree on the matching dimension according to the comparison result to obtain the component matching degree of the intelligent model and the execution terminal on the matching dimension.

5. The method according to claim 4, the method further comprises: When the execution terminal has a user authorization identifier, reading the configuration parameters of the terminal configuration item on the execution terminal; Wherein, the user authorization identifier indicates that the user to whom the execution terminal belongs at least allows the configuration parameters to be read.

6. The method according to claim 4, adjusting the initial matching degree on the matching dimension according to the comparison result, comprising: When the comparison result indicates that the demand parameter is superior to the configuration parameter, reducing the corresponding value of the initial matching degree on the matching dimension according to the difference degree characterized by the comparison result; When the comparison result indicates that the demand parameter is inferior to the configuration parameter, increasing the corresponding value of the initial matching degree on the matching dimension according to the difference degree characterized by the comparison result; When the comparison result indicates that the demand parameter is consistent with the configuration parameter, maintaining the initial matching degree on the matching dimension.

7. The method according to claim 1 or 2, wherein the intelligent model is determined by the following method: Obtaining the meeting type of the meeting to which the execution terminal belongs; According to the meeting type, screen for candidate models that match the meeting type in a model library containing multiple candidate models; Process the candidate models to match the execution terminal to obtain the intelligent model.

8. The method according to claim 7, wherein the intelligent model matches the meeting type, including: The intelligent model can execute any meeting tasks involved in the meeting type; Among them, processing the candidate models to match the execution terminal to obtain the intelligent model includes: Adjust the model parameters of the candidate models so that the data processing accuracy of the candidate models matches the execution capabilities of the execution terminal.

9. The method according to claim 1 or 2, according to the target processing intention and the description files of multiple intelligent models, call the target model to obtain the processing result output by the target model, including: Provide the target processing intention and the description files of multiple intelligent models to an inference model, and the inference model calls the target model that matches the target processing intention according to the description files through an execution function, so that the target model outputs a processing result.

10. An execution device for an intelligent model, comprising: An intention acquisition unit, configured to acquire a target processing intention in response to a processing request sent by any execution terminal; A model call unit, configured to call a target model according to the target processing intention and the description files of multiple intelligent models to obtain the processing result output by the target model; Among them, the target model is a model that matches the target processing intention among the multiple intelligent models; the description file of the intelligent model at least characterizes the execution terminal on which the intelligent model is installed and the functions that the intelligent model can implement; the execution terminal on which the intelligent model is installed is determined based on the device matching degree between the intelligent model and the execution terminal.

Images