CN118034913A - Cloud cooperative control method, electronic equipment and integrated large model deployment architecture - Google Patents

Abstract

The invention provides a cloud cooperative control method, electronic equipment and an integrated large model deployment architecture, and relates to the technical field of artificial intelligence, wherein the method comprises the following steps: acquiring the resource use condition of each terminal; generating a terminal calculation force condition table according to the resource use condition; and respectively deploying the models with various scales to each terminal according to the terminal calculation condition table, wherein the model of the terminal is used for performing first-stage task processing according to the original data, and the task processing result of the first stage is used for performing model reasoning in cooperation with the model deployed by the edge equipment or the model deployed by the cloud. The invention has the beneficial effects that: and the computing power resource is fully utilized and the privacy of the user is protected when model reasoning is carried out.

Description

Cloud cooperative control method, electronic equipment and integrated large model deployment architecture

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a cloud cooperative control method, electronic equipment and an integrated large model deployment architecture.

Background

As artificial intelligence progresses more and more rapidly, an integrated large model is formed by integrating a plurality of sub-models into a unified model, so that the artificial intelligence system has wider application scenes. In the integrated large model, each sub-model is responsible for processing specific tasks or functions so as to simultaneously process a plurality of tasks, and each sub-model cooperates without independently calling a plurality of models, so that a large model reasoning task can be finished quickly finally.

The usage scenarios of the integrated large model include, but are not limited to, the following: 1) Multitasking learning: i.e. simultaneously process a plurality of tasks, and improve the efficiency and performance of the model. 2) Joint training: the method is characterized in that data and features of a plurality of tasks are shared and learned, so that generalization capability of a model is improved, and for example, an integrated large model can simultaneously perform tasks such as target detection, image segmentation and attitude estimation. 3) Resource sharing: i.e. by sharing model parameters and computing resources, the storage and computation overhead of the model is reduced, which is very important for resource-constrained devices and environments.

However, when the integrated large model is deployed, the problems of limited terminal resources, serious dependence of cloud deployment network, insufficient security privacy protection and the like exist.

Disclosure of Invention

The invention solves the problem of how to fully utilize the computational power resources and protect the privacy of users when model reasoning is carried out.

In order to solve the problems, the invention provides a cloud cooperative control method, which comprises the following steps:

acquiring the resource use condition of each terminal;

Generating a terminal calculation force condition table according to the resource use condition;

And respectively deploying the models with various scales to each terminal according to the terminal calculation condition table, wherein the models of the terminals are used for performing first-stage task processing according to the original data, and the task processing results of the first stage are used for performing model reasoning in cooperation with the models deployed by the edge equipment or the models deployed by the cloud.

According to the cloud cooperative control method, the computing power resources of each terminal are reflected by the terminal computing power condition table, so that the models matched with the computing power condition table are deployed at the terminal, meanwhile, the models are deployed at the edge equipment and the cloud end, so that the distributed deployment of the integrated large model is formed, when the user needs artificial intelligent service, the initial model reasoning acceleration is firstly carried out based on the models of the terminal, and then the model reasoning results are sent to the edge equipment or the cloud end for carrying out the subsequent model reasoning according to the situation, so that the computing power of the terminal and the edge equipment is fully utilized, the cloud end burden is reduced, and at the moment, the cloud end server can be more focused on more complex generating tasks, and the cost of enterprises is reduced. Because the edge equipment is introduced to carry out the deployment of the model and the model reasoning at the subsequent stage, a more lightweight model can be deployed at the terminal, the large-scale model distribution deployment of the terminal is not required to be completed in a 'stacking force' mode, and the terminal can have more calculation force to carry out local tasks, so that the resource waste is reduced. In addition, the terminal transmits incomplete original data to the edge device and the cloud, and the original data can be selectively reserved to the terminal at the moment, so that the safety of user data is improved.

Further, the deploying the model of multiple scales to each terminal according to the terminal calculation force condition table includes:

And sending the terminal calculation condition table to the cloud so that the cloud deploys the models with various scales to the terminal respectively, wherein when the cloud identifies the terminal calculation condition table matched with the terminal, the model matched with the calculation scale of the terminal calculation condition table is sent to the terminal, and when the cloud does not identify the terminal calculation condition table matched with the terminal, the model with the pre-imputation force scale is sent to the terminal.

Further, the cloud cooperative control method further comprises the steps of:

acquiring an edge equipment identification request from the terminal, wherein the edge equipment identification request is generated according to the task processing result in the first stage;

determining whether the edge equipment meeting the task requirement exists according to the edge equipment identification request and an edge equipment routing table;

When the edge equipment meeting the task requirement exists, the identification information of the edge equipment meeting the task requirement is sent to the terminal, so that the terminal sends the task processing result of the first stage to the edge equipment meeting the task requirement.

Further, the edge device routing table comprises an edge device model deployment condition and an edge device resource use condition; the step of determining whether the edge device meeting the task requirement exists according to the edge device identification request and an edge device routing table comprises the following steps:

determining the model reasoning requirement of the terminal according to the edge equipment identification request;

and determining whether the edge equipment meets the task requirement according to the model reasoning requirement, the edge equipment model deployment condition and the edge equipment resource use condition.

Further, the edge device routing table further includes edge device location information; the sending the identification information of the edge device meeting the task requirement to the terminal includes:

and sending the identification information of the edge equipment closest to the terminal position in the edge equipment meeting the task requirement to the terminal according to the edge equipment position information.

Further, the cloud cooperative control method further comprises the following steps:

and when the edge equipment meeting the task requirement does not exist, sending a preset notification to the terminal so that the terminal can send the task processing result in the first stage to the cloud.

Further, the edge device is configured to generate a task processing result of a second stage according to the task processing result of the first stage and the model deployed by the edge device, and send the task processing result of the second stage to the cloud; the cloud end is used for generating a task processing result of a third stage according to the task processing result of the second stage and the model deployed by the cloud end, and sending the task processing result of the third stage to the terminal and/or the edge equipment.

The invention also provides electronic equipment, which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for realizing the cloud cooperative control method when executing the computer program.

The electronic device has similar technical effects to those of the cloud cooperative control method, and detailed description is omitted here.

The invention also provides an integrated large model deployment architecture, which comprises the electronic equipment.

Further, the integrated large model deployment architecture further comprises:

the terminal is used for generating a resource use condition, sending the resource use condition to the electronic equipment, deploying a model issued from a cloud, performing task processing at a first stage according to the model deployed by the terminal and original data, and sending a task processing result at the first stage to edge equipment or the cloud;

The edge equipment is used for receiving the task processing result from the first stage of the terminal so as to perform model reasoning according to the model deployed by the edge equipment and the task processing result of the first stage;

The cloud end is used for deploying the models of various scales to the terminal according to a terminal computing power situation table from the electronic equipment, and receiving the task processing result of the first stage from the terminal or the model reasoning result of the edge equipment so as to conduct model reasoning according to the cloud end deployed model and the task processing result of the first stage or the model reasoning result of the edge equipment.

The integrated large model deployment architecture has similar technical effects to the cloud cooperative control method and the electronic equipment, and detailed description is omitted here.

Drawings

Fig. 1 is a flowchart of a cloud cooperative control method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a model deployment of a vehicle end, cloud end and road side edge device according to an embodiment of the present invention;

FIG. 3 is a second schematic diagram of model deployment of a vehicle end, cloud end, and road side edge device according to an embodiment of the present invention;

fig. 4 is a schematic diagram of a model deployment of a vehicle end, a cloud end, and a road side edge device according to an embodiment of the present invention;

fig. 5 is a second flowchart of a cloud cooperative control method according to an embodiment of the present invention;

Fig. 6 is a flowchart III of a cloud cooperative control method according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of an integrated large model according to an embodiment of the present invention;

fig. 8 is a schematic flow chart of model reasoning among a vehicle end, a cloud end and a road side edge device according to an embodiment of the present invention;

fig. 9 is an application scenario schematic diagram of a cloud cooperative control method, an electronic device and an integrated large model deployment architecture according to an embodiment of the present invention;

fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. While the invention is susceptible of embodiment in the drawings, it is to be understood that the invention may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustration purposes only and are not intended to limit the scope of the present invention.

It should be understood that the various steps recited in the method embodiments of the present invention may be performed in a different order and/or performed in parallel. Furthermore, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the invention is not limited in this respect.

The term "including" and variations thereof as used herein are intended to be open-ended, i.e., including, but not limited to. The term "based on" is based at least in part on. The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "alternative embodiments". Related definitions of other terms will be given in the description below. It should be noted that the terms "first," "second," and the like herein are merely used for distinguishing between different devices, modules, or units and not for limiting the order or interdependence of the functions performed by such devices, modules, or units.

Referring to fig. 1, the embodiment of the invention provides a cloud cooperative control method, which includes the steps:

acquiring the resource use condition of each terminal;

And generating a terminal calculation force condition table according to the resource use condition.

Referring to fig. 9, the cloud cooperative control method in the embodiment of the present invention may be applicable to various application scenarios, such as traffic fields, industrial internet, general entertainment, smart medical fields, etc., for example, in traffic fields, may be applicable to specific scenarios such as V2X (Vehicle to X), automatic driving, etc. In this regard, the terminals may include a car end, ioT (Internet of Things ), a cell phone end, other device ends, and the like.

The resource usage of the terminal may be a resource usage of a processor of the terminal, for indicating a current computing power of the terminal. Finally, the resource use conditions of the terminals are respectively stored and summarized as a terminal calculation power condition table. In one embodiment, the resource usage of the terminal may be acquired in regions, for example, the cloud cooperative control method in the embodiment of the present invention is applied to an electronic device, where the terminal is a vehicle end, and the electronic device is responsible for acquiring the resource usage of the vehicle end in a road region.

And respectively deploying the models with various scales to each terminal according to the terminal calculation condition table, wherein the models of the terminals are used for performing first-stage task processing according to the original data, and the task processing results of the first stage are used for performing model reasoning in cooperation with the models deployed by the edge equipment or the models deployed by the cloud.

In the embodiment of the invention, each part of the integrated large model is deployed by using a terminal, an edge device and a cloud end respectively.

For the terminal, the calculation power is limited, so that when the model deployment is carried out, the model scale of the terminal deployment is selected by acquiring a terminal calculation power condition table capable of representing the calculation power of the terminal, for example, the terminal is a vehicle end, the current calculation power resource of the vehicle A is less, the model scale of the transmitted and deployed model is relatively smaller, and conversely, the current calculation power resource of the vehicle B is sufficient, and the model scale of cloud transmission is relatively larger.

For edge devices, particularly including MEC (Multi-ACCESS EDGE Computing) devices, which are Computing devices located at the edge of a network, to provide low-latency, high-bandwidth Computing and storage capabilities, the MEC devices are typically deployed on base stations, edge servers, or edge nodes at the edge of the network, and can cooperate with cloud servers to provide fast-response Computing services for end users. In an integrated large model deployment scenario, MEC devices may be used for communication and data transmission with cloud servers. For example, in the deployment of a large-scale machine learning model, preprocessing and feature extraction may be performed by the MEC device, and then the processed data is transmitted to a cloud server for model training and reasoning. The MEC device can provide low-delay computing and storage capacity, reduce delay and bandwidth consumption of data transmission, and improve efficiency and performance of model deployment. Therefore, the edge equipment can deploy the sub-model of the integrated model in real time or in advance according to the actual computing power condition, and in general, the computing power resource is larger than the terminal and smaller than the cloud, so that the model which can be deployed is moderate in scale.

For the cloud end, such as a server end of a manufacturer, the cloud end has the greatest calculation power, so that the cloud end can be deployed with the largest-scale integrated large model for processing data processed by the terminal model or the edge equipment model, or can directly process the most original data sent by the terminal.

Taking a terminal as a vehicle end and an edge device as a road side edge device as an example, referring to fig. 2, the invention is an integrated large model deployment scheme in an embodiment of the invention, wherein the vehicle end, the road side edge device and a cloud end are respectively interconnected, and cloud end computing power and resources > road side edge device computing power and resources > vehicle end computing power and resources, so that the vehicle end deploys a minimum-scale and lightest model according to the computing power, and when a user uses an AI service, the processing and response of a first stage can be completed locally and rapidly at the vehicle end. One or more models can be deployed on the road side edge equipment with resources inferior to the cloud end to finish the second step of processing, and then the model with larger resources is deployed on the cloud end to finish the third step of processing on the cloud end. For example, the current integrated large model data processing flow is: image classification- (image segmentation- (situation estimation-) simulation deduction). The image classification model can be deployed at the vehicle end, the classification of the pictures can be completed locally and rapidly, and then the result is transmitted to the road side edge equipment. The image segmentation model and the situation estimation model can be deployed on the road side edge equipment and are responsible for completing the second stage of processing, namely segmentation and situation estimation, and then the result is transmitted to the cloud to complete the third stage of processing. The cloud end deploys the simulation deduction large model with the heaviest resource occupation ratio and is responsible for completing deduction work in the third stage, and through the splitting and scheduling method, the integrated large model is moved from the cloud end to a user, and meanwhile, the acceleration of large model reasoning is completed.

It can be understood that in some cases, the model result obtained by reasoning the model deployed by the edge device according to the task processing result in the first stage is a final result, and the final result can be directly fed back to the terminal without being sent to the cloud. In addition, in some cases, the model deployed by the edge device may not meet the task requirement of the terminal, so that the terminal may directly send the task processing result in the first stage to the cloud end, so as to utilize the model deployed by the cloud end to perform model reasoning.

In summary, according to the cloud cooperative control method in the embodiment of the invention, by acquiring the computing power resource condition of the terminal, the models matched with the computing power resource condition are deployed at the terminal, and meanwhile, the models are deployed at the edge equipment and the cloud end, so that the distributed deployment of the integrated large model is formed, when the user needs artificial intelligent service, the initial model reasoning acceleration is firstly carried out based on the model of the terminal, and then the model reasoning result is sent to the edge equipment or the cloud end for carrying out the subsequent model reasoning according to the condition, so that the computing power of the terminal and the edge equipment is fully utilized, the cloud end burden is reduced, and at the moment, the cloud end server can be more focused on more complex generating tasks, and the cost of enterprises is reduced. Because the edge equipment is introduced to carry out the deployment of the model and the model reasoning at the subsequent stage, a more lightweight model can be deployed at the terminal, the large-scale model distribution deployment of the terminal is not required to be completed in a 'stacking force' mode, and the terminal can have more calculation force to carry out local tasks, so that the resource waste is reduced. In addition, the terminal transmits incomplete original data to the edge device and the cloud, and the original data can be selectively reserved to the terminal at the moment, so that the safety of user data is improved.

The scale referred to in the embodiment of the present invention may refer to the size of the computing power or the architecture in one model, or may be the size in number, for example, when the computing power resource of the terminal is more, it may deploy more models.

In an optional embodiment of the invention, the deploying models of multiple scales to each of the terminals according to the terminal calculation case table includes:

And sending the terminal calculation condition table to the cloud so that the cloud deploys the models with various scales to the terminal respectively, wherein when the cloud identifies the terminal calculation condition table matched with the terminal, the model matched with the calculation scale of the terminal calculation condition table is sent to the terminal, and when the cloud does not identify the terminal calculation condition table matched with the terminal, the model with the pre-imputation force scale is sent to the terminal.

Referring to fig. 3 to 5, the cloud cooperative control method in the embodiment of the invention can be applied to electronic equipment, and specifically is a reinforcement learning scheduler. Taking a terminal as a vehicle end as an example, the vehicle end of an area actively reports real-time use conditions (how often a specific interval is sent and can be manually configured) of respective resources to a scheduler by using a heartbeat mechanism, so that the scheduler is stored as a terminal calculation condition table and maintained, when a cloud end needs to issue deployment or update a model to the vehicle end, the terminal calculation condition table of the scheduler is checked first, then models with different scales are sent according to calculation conditions of the vehicle end, at this moment, the cloud end can judge whether a terminal calculation condition table matched with the vehicle end exists in the terminal calculation condition table sent by the scheduler according to information of the vehicle end, such as an ID (identity) of the vehicle end, if so, the second step is continued, the vehicle end issues the model with the corresponding scale according to the terminal calculation of the vehicle end reflected by the terminal calculation condition table, and then the vehicle end receives the model and performs deployment or update. In some cases, the terminal calculation situation table reported by the scheduler may not correspond to the vehicle end in the area, that is, when the cloud end cannot identify the terminal calculation situation table matched with the vehicle end, the minimum model (the model of the preset calculation scale) can be issued to the vehicle end uniformly, and meanwhile, the scheduler can be informed to collect the calculation situation of the vehicle end, update the terminal calculation situation table in time and perform deployment and update of the subsequent vehicle end model.

In an alternative embodiment of the present invention, the cloud cooperative control method further includes the steps of:

acquiring an edge equipment identification request from the terminal, wherein the edge equipment identification request is generated according to the task processing result in the first stage;

determining whether the edge equipment meeting the task requirement exists according to the edge equipment identification request and an edge equipment routing table;

When the edge equipment meeting the task requirement exists, the identification information of the edge equipment meeting the task requirement is sent to the terminal, so that the terminal sends the task processing result of the first stage to the edge equipment meeting the task requirement.

Referring to fig. 6, after completing the task processing result of the first stage, the terminal preferentially selects the edge device to perform model reasoning of the subsequent stage, where the corresponding edge device needs to be requested to perform model processing of the subsequent stage, so that the terminal requests to allocate the corresponding edge device to the scheduler according to the task processing result of the first stage, that is, sends an edge device identification request to the scheduler, and subsequently, when the scheduler receives the edge identification request from the terminal, that is, determines whether there is an edge device meeting the task requirement of the terminal to enable model reasoning of the subsequent stage according to the edge device routing table stored by the scheduler, if there is an edge device meeting the task requirement, for example, sends the identification information of the edge device meeting the task requirement to the terminal, for example, sends the location information or the ID information of the terminal device to the terminal, so that the terminal can find the corresponding edge device according to the identification information, and send the task processing result of the first stage to the edge device, so as to use the model deployed by the edge device to perform model reasoning of the subsequent stage.

In an optional embodiment, the edge device is configured to generate the task processing result of the second stage according to the task processing result of the first stage and the model deployed by the edge device, and send the task processing result of the second stage to the cloud end; the cloud end is used for generating a task processing result of a third stage according to the task processing result of the second stage and the model deployed by the cloud end, and sending the task processing result of the third stage to the terminal and/or the edge equipment.

Referring to fig. 6, in the embodiment of the present invention, a terminal, an edge device and a cloud end respectively process model reasoning processes in different stages, specifically, a model deployed by the terminal performs model reasoning in a first stage, and at this time, an edge device meeting task requirements of model reasoning in a subsequent second stage (i.e., a situation that the next-hop edge device exists is determined in fig. 6) can be found, so that the terminal sends a task processing result in the first stage to the edge device, and the edge device completes processing in the second stage according to the deployed model, and sends a task processing result in the second stage to the cloud end, and then the cloud end completes model reasoning in a third stage according to the task processing result in the second stage. The final third stage task processing results are the content required by the end user. Therefore, the inference is sequentially carried out through the models of different scales deployed by the terminal, the edge equipment, the cloud end and the like, so that the computing power resource is fully utilized, and the inference speed is improved.

Finally, the task processing result of the third stage completed by the cloud can be directly sent to the terminal to be fed back to the user, and also can be sent to the edge equipment to be cached on the edge equipment, so that the subsequent terminal can directly feed back the result through the edge equipment when the same task is required, and the cache acceleration function is realized.

The edge device routing table is generated and updated according to the information of the edge device, in some cases, a dispatcher executing the cloud cooperative control method according to the embodiment of the present invention may not find a corresponding edge device to complete the model reasoning requirement of the subsequent stage of the terminal, at this time, when the edge device meeting the task requirement does not exist, a preset notification is sent to the terminal, so that the terminal sends the task processing result of the first stage to the cloud, and referring to fig. 6, the cloud directly processes a request initiated by the terminal (vehicle end), and the cloud directly completes complete reasoning based on the task processing result of the first stage sent by the terminal and feeds back the result to the terminal.

Referring to fig. 7, an integrated large model composition diagram is shown, which includes four models, namely a model 1-image preprocessing model, a model 2-classification model, a model 3-segmentation model and a model 4-situation model, wherein the model 4 can be generally deployed at a cloud end, the models 1-3 can be deployed at a terminal and an edge device according to conditions, for example, the models 1 and 2 are deployed at the terminal, and the model 3 is deployed at the edge device.

The original data is a picture, the processed output 1 is obtained after the preprocessing function of the model 1, and then the model executed in the next step is judged according to the current task requirement. If the current demand is to classify an image, branch 1 is selected, namely a model 2-classification model is used to classify and identify the picture sample; if the current requirement is image segmentation, a model 3-segmentation model which jumps to a branch II is used for segmenting the picture; if the situation estimation is needed for the person or object in the picture, the process jumps to the model 4-situation estimation model to finish the task.

Referring to fig. 8, taking a terminal as an example, a model deployed at the vehicle end is an image classification model, a model deployed at the road side edge device is a segmentation model, a situation deduction model is deployed at the cloud end, and the colors of the image classification model at the vehicle end, the segmentation model of the road side edge device and the situation deduction model at the cloud end are from light to deep, which indicates that the scale and the weight of the model are sequentially increased. Outputting a task processing result in the first stage, namely outputting an image classification model, outputting a task processing result in the second stage, namely outputting a segmentation model, and outputting a task processing result in the third stage, namely outputting a situation deduction model. When the vehicle end uses the integrated large model service, the first step is to finish the first stage processing locally at the vehicle end, and the second step is to send a request to the dispatcher, the dispatcher finds the edge equipment responsible for the second stage processing according to the edge equipment routing table, and the identification information of the edge equipment is returned to the vehicle end. And thirdly, the vehicle end sends the task processing result of the first stage to the edge equipment according to the identification information of the edge equipment returned by the dispatcher. And fourthly, after the edge equipment finishes the second stage of processing, sending the result to the cloud end, and finishing the third stage of processing by the cloud end.

In an optional embodiment of the present invention, the edge device routing table includes an edge device model deployment scenario and an edge device resource usage scenario; the step of determining whether the edge device meeting the task requirement exists according to the edge device identification request and an edge device routing table comprises the following steps:

determining the model reasoning requirement of the terminal according to the edge equipment identification request;

and determining whether the edge equipment meets the task requirement according to the model reasoning requirement, the edge equipment model deployment condition and the edge equipment resource use condition.

In the embodiment of the invention, the edge device identification request is generated according to the task processing result of the first stage, for example, the task processing result of the first stage is image classification, if the task processing of the second stage is to be performed, an image segmentation model is required, a certain computing power is required, other computing power such as a language model cannot be completed, and the requirement cannot be met when the computing power of the edge device is smaller, so that the edge device identification request can embody the subsequent model reasoning requirement of the terminal, the edge device routing table comprises information of each edge device, namely, the information comprises model deployment condition and edge device resource use condition, and the information can be actively reported to the dispatcher by the edge device through a heartbeat mechanism, so that after the model reasoning requirement is determined according to the edge device identification request, the edge device model deployment condition and the edge device resource use condition in the edge device routing table can be combined to judge whether the edge device meeting the task requirement exists.

In addition, the edge device routing table also comprises edge device position information; the sending the identification information of the edge device meeting the task requirement to the terminal includes:

and sending the identification information of the edge equipment closest to the terminal position in the edge equipment meeting the task requirement to the terminal according to the edge equipment position information.

If the number of the edge devices meeting the task demand is determined in the edge device routing table, the edge device closest to the terminal can be determined at the moment, the identification information of the edge device is sent to the terminal, so that the terminal can correspondingly send the task processing result in the first stage through the identification information to complete the subsequent model reasoning, further the task processing result is obtained more rapidly,

The identification information may be edge device location information, that is, the terminal determines, according to the edge device location information, the edge device that receives the data of the edge device, and then sends a task processing result in the first stage. In other embodiments, the representation information may also be ID information of the edge device, whereby the edge device paired with it is determined by a specific ID.

In another embodiment of the present invention, an electronic device includes a memory for storing a computer program and a processor for implementing the cloud cooperative control method as described above when executing the computer program.

Referring to fig. 10, the electronic device includes a Central Processing Unit (CPU) 301 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 302 or a program loaded from a storage section 308 into a Random Access Memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic apparatus 300 are also stored. The CPU 301, ROM 302, and RAM 303 are connected to each other through a bus 304. An input/output (I/O) interface 305 is also connected to bus 304. In some embodiments, the following components are connected to the I/O interface 305: an input section 306 including a keyboard, a mouse, and the like; an output portion 307 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, a speaker, and the like; a storage section 308 including a hard disk or the like; and a communication section 309 including a network interface card such as a LAN card, a modem, or the like. The communication section 309 performs communication processing via a network such as the internet. The drive 310 is also connected to the I/O interface 305 as needed. A removable medium 311 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed on the drive 310 as needed, so that a computer program read therefrom is installed into the storage section 308 as needed.

The electronic device has similar technical effects to those of the cloud cooperative control method, and detailed description is omitted here.

An integrated large model deployment architecture of another embodiment of the present invention includes an electronic device as described above.

In an alternative embodiment of the invention, the integrated large model deployment architecture further comprises:

the terminal is used for generating a resource use condition, sending the resource use condition to the electronic equipment, deploying a model issued from a cloud, performing task processing at a first stage according to the model deployed by the terminal and original data, and sending a task processing result at the first stage to edge equipment or the cloud;

The edge equipment is used for receiving the task processing result from the first stage of the terminal so as to perform model reasoning according to the model deployed by the edge equipment and the task processing result of the first stage;

The cloud end is used for deploying the models of various scales to the terminal according to a terminal computing power situation table from the electronic equipment, and receiving the task processing result of the first stage from the terminal or the model reasoning result of the edge equipment so as to conduct model reasoning according to the cloud end deployed model and the task processing result of the first stage or the model reasoning result of the edge equipment.

Referring to fig. 4, the integrated large model deployment architecture in the embodiment of the invention can be specifically applied to the road traffic field, a terminal is a vehicle terminal, and an edge device is a road side edge device. The scheduler may use reinforcement learning training to take care of scheduling to reduce the cost of manual design and maintenance.

The integrated large model deployment architecture has similar technical effects to the cloud cooperative control method and the electronic equipment, and detailed description is omitted here.

Although the invention is disclosed above, the scope of the invention is not limited thereto. Various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the invention, and such changes and modifications would be within the scope of the invention.

Claims (10)

1. The cloud cooperative control method is characterized by comprising the following steps of:

acquiring the resource use condition of each terminal;

Generating a terminal calculation force condition table according to the resource use condition;

And respectively deploying the models with various scales to each terminal according to the terminal calculation condition table, wherein the models of the terminals are used for performing first-stage task processing according to the original data, and the task processing results of the first stage are used for performing model reasoning in cooperation with the models deployed by the edge equipment or the models deployed by the cloud.

2. The cloud cooperative control method according to claim 1, wherein the deploying models of a plurality of scales to each of the terminals according to the terminal calculation condition table includes:

And sending the terminal calculation condition table to the cloud so that the cloud deploys the models with various scales to the terminal respectively, wherein when the cloud identifies the terminal calculation condition table matched with the terminal, the model matched with the calculation scale of the terminal calculation condition table is sent to the terminal, and when the cloud does not identify the terminal calculation condition table matched with the terminal, the model with the pre-imputation force scale is sent to the terminal.

3. The cloud cooperative control method of claim 1, further comprising:

acquiring an edge equipment identification request from the terminal, wherein the edge equipment identification request is generated according to the task processing result in the first stage;

determining whether the edge equipment meeting the task requirement exists according to the edge equipment identification request and an edge equipment routing table;

When the edge equipment meeting the task requirement exists, the identification information of the edge equipment meeting the task requirement is sent to the terminal, so that the terminal sends the task processing result of the first stage to the edge equipment meeting the task requirement.

4. The cloud cooperative control method of claim 3, wherein the edge device routing table includes an edge device model deployment scenario and an edge device resource usage scenario; the determining whether the edge device meeting the task requirement exists according to the edge device identification request and an edge device routing table comprises the following steps:

determining the model reasoning requirement of the terminal according to the edge equipment identification request;

and determining whether the edge equipment meets the task requirement according to the model reasoning requirement, the edge equipment model deployment condition and the edge equipment resource use condition.

5. The cloud cooperative control method of claim 4, wherein the edge device routing table further comprises edge device location information; the sending the identification information of the edge device meeting the task requirement to the terminal includes:

and sending the identification information of the edge equipment closest to the terminal position in the edge equipment meeting the task requirement to the terminal according to the edge equipment position information.

6. The cloud cooperative control method of claim 3, further comprising:

and when the edge equipment meeting the task requirement does not exist, sending a preset notification to the terminal so that the terminal can send the task processing result in the first stage to the cloud.

7. The cloud cooperative control method according to claim 3, wherein the edge device is configured to generate the task processing result of a second stage according to the task processing result of the first stage and the model deployed by the edge device, and send the task processing result of the second stage to the cloud; the cloud end is used for generating a task processing result of a third stage according to the task processing result of the second stage and the model deployed by the cloud end, and sending the task processing result of the third stage to the terminal and/or the edge equipment.

8. An electronic device comprising a memory for storing a computer program and a processor for implementing the cloud cooperative control method according to any of claims 1 to 7 when the computer program is executed.

9. An integrated large model deployment architecture comprising the electronic device of claim 8.

10. The integrated large model deployment architecture of claim 9, further comprising:

the terminal is used for generating a resource use condition, sending the resource use condition to the electronic equipment, deploying a model issued from a cloud, performing task processing at a first stage according to the model deployed by the terminal and original data, and sending a task processing result at the first stage to edge equipment or the cloud;

The edge equipment is used for receiving the task processing result from the first stage of the terminal so as to perform model reasoning according to the model deployed by the edge equipment and the task processing result of the first stage;

The cloud end is used for deploying the models of various scales to the terminal according to a terminal computing power situation table from the electronic equipment, and receiving the task processing result of the first stage from the terminal or the model reasoning result of the edge equipment so as to conduct model reasoning according to the cloud end deployed model and the task processing result of the first stage or the model reasoning result of the edge equipment.