Method, system and cluster for generating content by using model

Abstract

The invention discloses a method, system and cluster for generating content by using a model, which can be applied to the technical field of cloud computing. The method comprises the steps that a client receives a content generation instruction sent by a user, wherein the content generation instruction is used for instructing a model to generate target content according to the intention of the user; the client inputs the content generation instruction to the first module, so that the first module outputs an intermediate result of model generation target content; the server receives the intermediate result sent by the client and inputs the intermediate result to the second module, so that the second module outputs the target content; and the server sends the target content to the client. The method can prevent the cloud from acquiring the user sensitive information in the content generation instruction, and protects the information security of the user.

Description

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office of China on June 28, 2023, with application number 202310777021.3 and application name “A Command Transmission Method”, the entire contents of which are incorporated by reference in this application. Technical Field

[0002] The present application relates to the technical fields of cloud computing and artificial intelligence (AI), and in particular to a method, system and cluster for generating content using a model. Background Art

[0003] AI-generated content (AIGC) is a technology that generates digital content (such as pictures, audio, and text) through AI models. Among them, the AI ​​model used to generate digital content can be called a generative model (GM) model. With the development of generative models, ordinary users can generate high-quality content through AIGC. Therefore, AIGC has been rapidly applied.

[0004] like Figure 1 As shown in the figure, since the generative model has high computing power requirements, the generative model is usually deployed in the cloud. As a result, users can use the generative model to generate content using lightweight terminal devices. Specifically, the user uses the terminal device to send content generation instructions to the generative model in the cloud. The generative model generates content that the user may want based on the content generation instructions, that is, the target content.

[0005] Content generation instructions may contain sensitive information of users. For example, users may accidentally include personal information, business secrets, or information reflecting user preferences in content generation instructions. Therefore, sending content generation instructions to the cloud may allow the cloud to obtain sensitive information of users, resulting in leakage of sensitive information of users. Summary of the invention

[0006] The embodiments of the present application propose a method, system and cluster for generating content using a model, which can prevent the cloud from obtaining user sensitive information in content generation instructions and protect user information security.

[0007] In a first aspect, a method for generating content using a model is provided, which is applied to a content generation system, wherein the system includes a client and a server, the model includes a first module and a second module, the client and the first module are deployed locally on the user, the server and the second module are deployed in the cloud, and the output of the first module is the input of the second module; the method includes: the client receives a content generation instruction issued by the user, the content generation instruction is used to instruct the model to generate target content according to the user's intention; the client inputs the content generation instruction to the first module, so that the first module outputs an intermediate result of the model generating the target content; the server receives the intermediate result sent by the client, and inputs the intermediate result to the second module, so that the second module outputs the target content; the server sends the target content to the client.

[0008] This method deploys part of the model used to generate the target content locally on the user, and the other part on the cloud. When the client deployed locally on the user receives the content generation instruction, it can use the local deployed part of the model to process the content generation instruction and obtain an intermediate result that is difficult to use to reverse the content generation instruction. Then, the intermediate result is sent to the cloud, so that the cloud generates the target content based on the intermediate result, thereby avoiding the leakage of sensitive information in the content generation instruction without requiring the user to run all the models locally.

[0009] In one possible implementation, the model is a generative model including multiple network layers, and the multiple network layers are arranged sequentially in the data transmission direction of the generative model, wherein the first module includes at least one network layer among the multiple network layers, and the second module includes network layers other than at least one network layer among the multiple network layers.

[0010] In this implementation, part of the network layer of the generative model can be deployed locally on the user, and the other part of the network layer can be deployed in the cloud. The network layer deployed locally on the user can process the content generation instructions issued by the user to obtain the intermediate results of the generative model. Due to the unexplainability of the AI ​​model, it is difficult to infer the input of the model, that is, the content generation instructions, from the intermediate results. Therefore, after obtaining the content generation instructions, it is difficult for the cloud to obtain the user's content generation instructions, and it is difficult to obtain sensitive information that may be contained in the content generation instructions.

[0011] In one possible implementation, the content generation instruction includes the user's intention, the model includes an interpreter and a generator, the interpreter is used to extract the characteristics of the user's intention and obtain a coding vector, and the generator is used to generate the target content based on the coding vector; wherein, the first module includes the interpreter, and the second module includes the generator; or, the first module includes the first network layer in the interpreter, and the second module includes the second network layer in the interpreter and the generator; the output of the first network layer is the input of the second network layer; or, the first module includes the interpreter and the third network layer in the generator, and the second module includes the fourth network layer in the generator; the output of the third network layer is the input of the fourth network layer.

[0012] In this implementation, the first module includes an interpreter, a part of the interpreter, or all of the interpreters and a part of the generator, that is, the first module can be flexibly configured. Among them, when the user's local computing power is high, the first module can include all of the interpreters, or all of the interpreters and a part of the generator, so that the user's local computing power can be fully utilized, and the intermediate results output by the first module are also more difficult to be used to infer content generation instructions. When the user's local computing power is low, the first module can include part of the interpreter, thereby reducing the demand for local computing power.

[0013] In a possible implementation, the model is a generative model, and the method further includes: the server trains the model; after the model training is completed, the server sends the first module to the user locally.

[0014] In this implementation, the model can be trained in the cloud, and after the model training is completed, the first module in the model is sent to the user locally. Thus, the user does not need to train or design the first module, which improves user experience.

[0015] In a possible implementation manner, the method further includes: the client receives a parameter adjustment operation issued by a user, and adjusts the parameters of the first module based on the parameter adjustment operation.

[0016] In this implementation, the user can adjust or modify the parameters of the first module so that the cloud or other methods cannot perceive the parameters of the first module and thus cannot use the parameters of the first module to infer the content generation instructions issued by the user.

[0017] In a possible implementation, the first module includes a sensitive information filter set by a user, and the filter is used to intercept sensitive information in the content generation instruction to obtain an intermediate result.

[0018] In this implementation, the first module can be implemented as a filter, which has low requirements on computing power, so that users can generate content through the method provided in the embodiment of the present application using a lightweight user local device.

[0019] In a second aspect, a content generation system is provided, which is used to generate content using a model, wherein the system includes a client and a server, the model includes a first module and a second module, the client and the first module are deployed locally on the user, the server and the second module are deployed in the cloud, and the output of the first module is the input of the second module; the client is used to receive a content generation instruction issued by the user, the content generation instruction is used to instruct the model to generate target content according to the user's intention; the client is used to input the content generation instruction into the first module, so that the first module outputs an intermediate result of the model generating the target content; the server is used to receive the intermediate result sent by the client, and input the intermediate result into the second module, so that the second module outputs the target content; the server is used to send the target content to the client.

[0020] In one possible implementation, the model is a generative model including multiple network layers, and the multiple network layers are arranged sequentially in the data transmission direction of the generative model, wherein the first module includes at least one network layer among the multiple network layers, and the second module includes network layers other than at least one network layer among the multiple network layers.

[0021] In one possible implementation, the content generation instruction includes the user's intention, the model includes an interpreter and a generator, the interpreter is used to extract the characteristics of the user's intention and obtain a coding vector, and the generator is used to generate the target content based on the coding vector; wherein, the first module includes the interpreter, and the second module includes the generator; or, the first module includes the first network layer in the interpreter, and the second module includes the second network layer in the interpreter and the generator; the output of the first network layer is the input of the second network layer; or, the first module includes the interpreter and the third network layer in the generator, and the second module includes the fourth network layer in the generator; the output of the third network layer is the input of the fourth network layer.

[0022] In a possible implementation, the model is a generative model, and the server is also used to: train the model; after the model training is completed, send the first module to the user locally.

[0023] In a possible implementation manner, the client is further used to: receive a parameter adjustment operation issued by a user, and adjust the parameters of the first module based on the parameter adjustment operation.

[0024] In a possible implementation, the first module includes a sensitive information filter set by a user, and the filter is used to intercept sensitive information in the content generation instruction to obtain an intermediate result.

[0025] In a third aspect, a method for generating content using a model is provided, which is applied to a client deployed locally on a user, wherein a content generation system in which the client is located also includes a server deployed in the cloud, and the model includes a first module and a second module, wherein the first module is deployed locally on the user, and the second module is deployed in the cloud, and the output of the first module is the input of the second module; the method includes: the client receives a content generation instruction issued by the user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; the client inputs the content generation instruction into the first module, so that the first module outputs an intermediate result of the target content generated by the model; the client sends the intermediate result to the server; wherein the server is used to input the intermediate result into the second module, so that the second module outputs the target content; the client receives the target content sent by the server.

[0026] In a possible implementation manner, the method further includes: the client receives a parameter adjustment operation issued by a user, and adjusts the parameters of the first module based on the parameter adjustment operation.

[0027] In a fourth aspect, a client for generating content using a model is provided. The client is deployed locally on the user, and a content generation system in which the client is located also includes a server deployed in the cloud. The model includes a first module and a second module. The first module is deployed locally on the user, and the second module is deployed in the cloud, and the output of the first module is the input of the second module. The client includes: a receiving module for receiving a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; an input module for inputting the content generation instruction into the first module, so that the first module outputs an intermediate result of the target content generated by the model; a communication module for sending the intermediate result to the server; wherein the server is used to input the intermediate result into the second module, so that the second module outputs the target content; the communication module is also used to receive the target content sent by the server.

[0028] In a possible implementation, the receiving module is further used to: receive a parameter adjustment operation issued by a user, and adjust the parameters of the first module based on the parameter adjustment operation.

[0029] In a fifth aspect, a computing device cluster is provided, comprising at least one computing device, each computing device comprising a processor and a memory; the processor of at least one computing device is used to execute instructions stored in the memory of at least one computing device, so that the computing device cluster executes the method provided in the third aspect.

[0030] In a sixth aspect, a computer-readable storage medium is provided, comprising computer program instructions. When the computer program instructions are executed by a computing device cluster, the computing device cluster executes the method provided in the third aspect.

[0031] In a seventh aspect, a computer program product comprising instructions is provided. When the instructions are executed by a computer device cluster, the computer device cluster executes the method provided in the third aspect.

[0032] The beneficial effects of the second to seventh aspects can be referred to the above introduction to the beneficial effects of the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of a content generation scheme;

[0034] Figure 2 It is a schematic diagram of the structure of a generative model;

[0035] Figure 3A A schematic diagram of a system architecture provided for an embodiment of the present application;

[0036] Figure 3B A schematic diagram of a system architecture provided for an embodiment of the present application;

[0037] Figure 3C A schematic diagram of a system architecture provided for an embodiment of the present application;

[0038] Figure 4 A flowchart of a content generation method provided in an embodiment of the present application;

[0039] Figure 5 A schematic diagram of a module 210 and a module 220 provided in an embodiment of the present application;

[0040] Figure 6 A schematic diagram of a module 210 and a module 220 provided in an embodiment of the present application;

[0041] Figure 7 A schematic diagram of a module 210 and a module 220 provided in an embodiment of the present application;

[0042] Figure 8 A schematic diagram of a module 210 and a module 220 provided in an embodiment of the present application;

[0043] Fig. 9 A schematic diagram of the structure of a content generation system provided in an embodiment of the present application;

[0044] Fig.10 A flowchart of a content generation method provided in an embodiment of the present application;

[0045] Fig.11 A schematic diagram of the structure of a client provided in an embodiment of the present application;

[0046] Fig.12A schematic diagram of the structure of a computing device provided in an embodiment of the present application;

[0047] Fig.13 A schematic diagram of the structure of a computing device cluster provided in an embodiment of the present application;

[0048] Fig.14 A schematic diagram of the structure of a computing device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will describe the solution provided by the embodiment of the present application in conjunction with the accompanying drawings. In the embodiment of the present application, "multiple" refers to two or more than two.

[0050] To facilitate understanding of the solutions provided by the embodiments of the present application, some technical terms that may be involved in the embodiments of the present application are first introduced before the solutions provided by the embodiments of the present application are introduced.

[0051] Content generation instructions: refers to information or commands issued by the user to instruct the generative model to generate content according to the user's intention. Content generation instructions include user intent. User intent is used to describe the target content (i.e. the content to be generated). For example, if the user wants to generate a painting of a house with a chimney, the user intent can be "draw a house with a chimney".

[0052] Generative model: A neural network that understands user intent and generates content based on that intent. In other words, a generative model takes content generation instructions as input and outputs content that matches that user intent. Figure 2 As shown in , the generative model usually consists of an interpreter and a generator. The interpreter is used to understand the user intent contained in the content generation instruction, that is, to extract the features of the user intent and obtain the encoding vector of the user intent. The encoding vector of the user intent is also called the feature vector of the user intent. The generator is used to generate content that meets the user intent according to the encoding vector of the user intent, using the generator's own knowledge and information. Figure 2As shown in the figure, the interpreter and the generator are respectively composed of multiple network layers arranged in sequence. Among them, the network layer can be simply referred to as a layer, and the output of each layer is the input of the next layer of the layer. Each layer includes multiple neurons, each neuron uses parameters to calculate the data and outputs the calculation result. The calculation result is usually represented by a vector, which can be called an encoding vector. Among them, the parameters of the neurons are obtained through training, and the training of the model or the training of the neural network is essentially the training or adjustment of the parameters of the neurons. At present, typical generative models include text-to-text transfer transformer (T5), generative adversarial networks (GANs), variational autoencoders (VAEs), etc.

[0053] Cloud: A remote device, equipment or platform relative to the user's local area. Specifically, it can be one or more servers or computing instances deployed in the cloud, such as a data center, server cluster, etc.

[0054] User local: refers to the user-side device, equipment or platform, which can be the user's terminal device, or the edge device or computing instance deployed by the cloud vendor on the user side.

[0055] Data masking: Data masking is the process of transforming sensitive information (identity information, asset information, etc.) using masking rules to prevent sensitive information from being leaked and to achieve reliable protection of sensitive information. Sensitive information can also be called private information.

[0056] Deep learning: A type of machine learning technology based on deep neural network algorithms. Its main feature is the use of multiple nonlinear transformations to process and analyze data. Deep learning is mainly used in the perception, decision-making and content generation scenarios in the field of artificial intelligence, such as image recognition, speech recognition, natural language translation, computer games, content generation, etc.

[0057] Figure 3A A system architecture provided by an embodiment of the present application is shown. The system architecture includes a content generation system 100 and a model 200. The content generation system 100 is composed of a client 110 and a server 120, and the model 200 is composed of a module 210 and a module 220. Among them, the client 110 and the module 210 are deployed locally on the user, and the server 120 and the module 220 are deployed in the cloud.

[0058] The client 110 can receive a content generation instruction issued by a user, and send the content generation instruction to the module 210. The module 210 can perform data processing on the content generation instruction to obtain an intermediate result. Among them, the data processing includes data desensitization, that is, the intermediate result is the desensitized data. The client 110 can obtain the intermediate result and send the intermediate result to the server 120 through the network. The server 120 inputs the content generation instruction (i.e., the intermediate result) sent from the client 110 into the module 220. The module 220 can generate and output the target content based on the content generation instruction. The server 120 can obtain the target content and send the target content to the client 110 through the network. The client 110 can display or play the target content to the user.

[0059] In the embodiment of the present application, the intermediate result refers to the intermediate result of the target content generated by the model 200. Specifically, the process from the input of the content generation instruction into the model 200 to the output of the target content by the model 200 includes multiple processing steps. For example, the model includes multiple network layers, and the processing of data by each network layer is a processing step. Except for the last processing step in the multiple processing steps, the processing result of each processing step can be called an intermediate result.

[0060] In some embodiments, the client 110 or the server 120 may be implemented in software, for example, the client 110 may be an application or process running locally on the user, and the server 120 may be an application or process running locally on the user. In some embodiments, the client 110 or the server 120 may be implemented in hardware, for example, the client 110 may be a terminal device, and the server 120 may be a server, etc. The embodiments of the present application do not specifically limit the implementation of the client 110 and the server 120.

[0061] In some embodiments, see Figure 3B , model 200 can be a generative model composed of multiple network layers. The multiple network layers are arranged in sequence in the data transmission direction of the generative model, that is, the output of one network layer is the input of one network layer of the network layer. Module 210 includes at least one network layer among the multiple network layers, and module 220 includes network layers other than the at least one network layer among the multiple network layers. The output of the network layer forming the first module is the input of the network layer forming the second module. That is to say, in the data transmission direction in the generative model, module 220 is the next layer or layers of module 210. That is to say, in this embodiment, a part of the multiple network layers constituting the generative model are deployed locally to the user, and the other part is deployed to the cloud.

[0062] Among them, the data transmission direction in the generative model refers to the transmission direction or forward direction of data in the generative model from data input to the generative model to the generative model outputting data.

[0063] In an embodiment of the present application, the generative model may be a model for generating content from text to text, a model for generating content from text to image, a model for generating content from text to video, or a model for generating content from text to audio.

[0064] Module 210 takes content generation instructions as input and outputs a coding vector. As an intermediate processing result of the AI ​​model, the coding vector is difficult to be used to infer the input information of the AI ​​model. In other words, the coding vector is an intermediate result in the process of the generative model generating the target content based on the content generation instruction, and it is difficult to be used to infer the content generation instruction input to the generative model. Specifically, the generative model is an AI model obtained based on deep learning training and is unexplainable. Up to now, there is no technology that can use the intermediate results or outputs of the generative model to infer the input of the generative model. Therefore, in this embodiment, at the user's local, a part of the network layer in the generative model is used to convert the content generation instruction issued by the user into the intermediate result of the generative model, and the intermediate result is sent to the cloud, thereby avoiding the original content generation instruction from being sent directly to the cloud, thereby avoiding the leakage of sensitive information. In addition, in this embodiment, the content generation instruction is processed using the user's local computing resources to obtain the intermediate result, which is directly used as the input of the module 220 deployed in the cloud, so that the cloud does not need to process the content generation instruction, saving the computing resources of the cloud and improving the utilization rate of the user's local computing resources.

[0065] In some embodiments, see Figure 3C , model 200 may be composed of a filter and a generative model. Among them, the generative model is used as module 220. The filter is also called a privacy filter and can be set by the user. The filter is used to intercept or filter sensitive information in the content generation instruction to obtain an intermediate result. The desensitized result here is also called the desensitized content generation instruction. In one example, the filter includes a blacklist set by the user. The blacklist includes sensitive information listed by the user. If the content generation instruction includes information in the blacklist, the filter intercepts the information on the blacklist in the content generation instruction. In one example, the filter includes a whitelist set by the user. The whitelist includes non-sensitive information listed by the user. If the content generation instruction includes information outside the whitelist, the filter intercepts the information outside the whitelist in the content generation instruction. In this embodiment, there is no need to run a neural network with high computing power requirements locally on the user, which reduces the computing power requirements for the user locally.

[0066] The above example introduces a system architecture provided by an embodiment of the present application. Next, in combination with the system architecture, a content generation method provided by an embodiment of the present application is introduced. The method can be specifically executed by the content generation system 100. Figure 4 As shown, the method includes the following steps.

[0067] Step 401 : The client 110 receives a content generation instruction issued by a user, where the content generation instruction is used to instruct the model 200 to generate target content according to the user's intention.

[0068] The content generation instruction includes the user's intention. Taking text-to-image content generation as an example, if the user wants an image of a house with a chimney, he can issue a content generation instruction including "draw a picture of a house with a chimney". Among them, "draw a picture of a house with a chimney" is the user's intention.

[0069] Among them, Figure 3A As shown, the client 110 is located locally, and the user can input the content generation instruction to the client 110 locally, which ensures that the original content generation instruction does not leave the user's local area and avoids privacy leakage. The original content generation instruction refers to the content generation instruction issued by the user and not processed by the module 210 or the model 200.

[0070] In some embodiments, the client 110 may provide a user interface, and the user interface may be used by the user to input the content generation instruction. For example, the user interface may include an edit box, and the user may input the content generation instruction into the edit box.

[0071] In step 402 , the client 110 inputs the content generation instruction to the module 210 , so that the module 210 outputs an intermediate result of the model generating the target content.

[0072] like Figure 3A As shown, the client 110 and the module 210 are both located locally in the user. Therefore, the content generation instruction can be input into the module 210, that is, the content generation instruction is input into the model 200, without leaving the user's local location, thereby avoiding the leakage of user privacy information.

[0073] Module 210 can perform data desensitization processing on the content generation instruction to obtain and output an intermediate result. The intermediate result after desensitization processing removes or protects sensitive information in the content generation instruction.

[0074] In some embodiments, Figure 3BAs shown, model 200 is a generative model, and module 210 is at least one network layer in the generative model. The at least one network layer can process the content generation instruction to obtain an intermediate result of the generative model generating the target content based on the content generation instruction. The network layers other than the at least one network layer in the multiple network layers are deployed in the cloud and used as module 220.

[0075] In one example, the at least one network layer may include an input layer, and the output layer may embed the content generation instruction to obtain a vector representation of the content generation instruction, that is, an encoding vector. The encoding vector may be used as an intermediate result output by module 210.

[0076] In one example, the at least one network layer may further include an input layer and a feature extraction layer. The feature extraction layer may further calculate the encoding vector of the content generation instruction to extract features of the content generation instruction, such as extracting features of the user's intention in the content generation instruction, to obtain a feature vector of the content generation instruction. The feature vector belongs to the encoding vector, which is the encoding vector output by the at least one network layer. The feature vector may be used as an intermediate result output by module 210.

[0077] In one example, see Figure 5 The generative model as model 200 includes an interpreter and a generator. The interpreter is used to understand the user intent contained in the content generation instruction and obtain the encoding vector of the user intent. The generator is used to generate content that meets the user intent according to the encoding vector of the user intent and using the generator's own knowledge and information.

[0078] In this example, the interpreter is deployed locally on the user side as module 210. The generator is deployed in the cloud as module 220. The content generation instruction includes the user's intention, such as "draw a house with a chimney". The interpreter as module 210 can extract the features of the user's intention and obtain a feature vector of the user's intention. Among them, extracting the features of the user's intention is also called understanding the user's intention, and the feature vector of the user's intention can also be called an interpretation result vector. Module 210 can output the interpretation result vector. The interpretation result vector is used as the intermediate result obtained in step 402. Among them, the interpretation result vector belongs to the encoding vector, specifically the encoding vector output by the interpreter.

[0079] In one example, see Figure 6, as the generative model of model 200, includes an interpreter and a generator. Among them, the interpreter can be composed of multiple network layers, and the multiple network layers are arranged in sequence in the data transmission direction of the generative model. The first network layer in the interpreter can be deployed to the user's local as module 210. The second network layer and the generator in the interpreter can be deployed to the cloud as module 220. Among them, the first network layer is at least one network layer among the multiple network layers that constitute the interpreter, the second network layer is the network layer other than the first network layer among the multiple network layers, and the output of the first network layer is the input of the second network layer. Among them, when the first network layer is two or more network layers, the output of the first network layer refers to the output of the last network layer in the two or more network layers in the data transmission direction. When the second network layer is two or more network layers, the input of the second network layer refers to the input of the first network layer in the two or more network layers in the data transmission direction.

[0080] The first network layer of module 210 can extract the features of the user's intention and obtain an intermediate feature vector of the user's intention. The intermediate feature vector of the user's intention can also be called an intermediate interpretation result vector. Module 210 can output the intermediate interpretation result vector. The intermediate interpretation result vector is used as the intermediate result obtained in step 402. Among them, the intermediate interpretation result vector belongs to the encoding vector, specifically the encoding vector output by the first network layer.

[0081] In this example, the module 210 deployed locally on the user is a partial interpreter, which can complete part of the interpretation process of the content generation instruction locally on the user, and the remaining interpretation process and the entire content generation process are performed in the cloud, which can reduce the local computing power requirements for the user. For example, the interpreter can be set as a 100-layer neural network, and the module 210 can include the first 50 layers of the neural network. The last 50 layers of the neural network and the generator are deployed in the cloud. In this way, only the first 50 layers of data processing of the content generation instruction are required locally on the user, and the local computing power requirements for the user are low.

[0082] In one example, see Figure 7, as the generative model of model 200, includes an interpreter and a generator. Among them, the generator may be composed of multiple network layers, and the multiple network layers are arranged in sequence in the data transmission direction of the generative model. The third network layer and the interpreter in the generator can be deployed to the user's local as module 210. The fourth network layer in the generator can be deployed to the cloud as module 220. Among them, the third network layer is at least one network layer among the multiple network layers that constitute the generator, the fourth network layer is the network layer other than the third network layer among the multiple network layers, and the output of the third network layer is the input of the fourth network layer. Among them, when the third network layer is two or more network layers, the output of the third network layer refers to the output of the last network layer in the data transmission direction of the two or more network layers. When the fourth network layer is two or more network layers, the input of the fourth network layer refers to the input of the first network layer in the data transmission direction of the two or more network layers.

[0083] In this example, the module 210 deployed locally on the user includes an interpreter and a partial generator (i.e., the third network layer). The interpreter can extract the features of the user's intention and obtain a feature vector of the user's intention. The partial generator can complete the partial generation process based on the feature vector and obtain a partial generation result. The partial generation result is used as the intermediate result obtained in step 402. The representation form of the partial generation result is also a vector, that is, the partial generation result also belongs to a coding vector, which is the coding vector output by the third network layer.

[0084] In this example, the intermediate result sent from the user's local computer to the cloud is a partial generation result. The partial generation result is obtained by performing more operations on the content generation instruction. It is more difficult to reversely infer the content generation instruction based on the partial generation result, thereby effectively avoiding the leakage of sensitive information.

[0085] In some embodiments, Figure 3C As shown, module 210 can be a filter deployed locally on the user, and module 220 is a generative model deployed in the cloud. Module 210 can also be called a sensitive information filter, which is used to intercept or filter sensitive information in content generation instructions to obtain desensitized content generation instructions. The desensitized content generation instructions are used as the intermediate result obtained in step 402. In other words, module 210 can identify sensitive information in content generation instructions, and intercept or filter the sensitive information so that the sensitive information does not leave the user's local area. For example, Figure 8 As shown, the content generation instruction issued by the user can be set as "My house has a chimney, draw a house with a chimney". Among them, "My house has a chimney" includes the user's personal information, which is sensitive information. Module 210 processes the content generation instruction, and the desensitized content generation instruction obtained is "Draw a house with a chimney", which does not contain the user's sensitive information.

[0086] In one example, module 210 may be a whitelist or a blacklist set by a user. For details, please refer to the above introduction and will not be described again here.

[0087] Back to Figure 4 In step 403, the server 120 receives the intermediate result sent by the client 110, and inputs the intermediate result to the module 220, so that the second module outputs the target content.

[0088] The client 110 may send the intermediate result output by the module 210 to the server 120 located in the cloud through the network. For example, the client 110 may send the intermediate result to the server 120 through the hypertext transfer protocol secure (HTTPS) or other network protocols.

[0089] When receiving the intermediate result sent by the client 110, the server 120 may input the intermediate result into the module 220. The module 220 may obtain the target content based on the intermediate result.

[0090] When module 210 is an interpreter and module 220 is a generator, the intermediate result is an interpretation result vector output by the interpreter, and the generator can generate target content based on the interpretation result vector.

[0091] When module 210 is the first network layer in the interpreter and module 220 is the second network layer and generator of the interpreter, the intermediate result is an intermediate interpretation result vector. The second network layer can continue to understand the content generation instruction based on the intermediate interpretation result to obtain the interpretation result vector, and the generator can generate the target content based on the interpretation result vector.

[0092] When module 210 is the third network layer in the interpreter and the generator, and module 220 is the fourth network layer in the generator, the intermediate result is a partial generation result. Module 220 can continue to generate content based on the partial generation result to obtain the target content.

[0093] When module 210 is a filter and module 220 is a generative model, the intermediate result is an instruction for generating content after desensitization. The generative model can generate target content based on the instruction for generating content after desensitization.

[0094] After generating the target content, the module 220 may output the target content, wherein the target content may be any one or more of pictures, videos, texts, and audios.

[0095] In step 404, the server 120 may send the target content output by the module 220 to the client 110. The server 120 may send the target content to the client 110 via HTTPS or other network protocols.

[0096] The client 110 may provide the target content to the user, for example, by displaying or playing the target content, so that the user obtains the target content.

[0097] In some embodiments, model 200 may be a generative model. Generative models are AI models that require training to obtain. The training of AI models requires high computing power and professional operations. Model 200 can be trained in the cloud. In other words, the training of model 200 can be performed by a cloud vendor. Among them, module 200 can be trained by server 120. After the training of model 200 is completed, that is, after the value of the loss function of model 200 is relatively small (for example, less than a set value), server 220 can obtain module 210 in model 200 and send module 210 to the user locally, thereby deploying module 210 locally.

[0098] In one example, the server 210 may provide a display interface for displaying that the module 210 is downloadable. The user may send a download request to the server 210 through the client 110. The server 210 may respond to the download request and send the module 210 to the user's local computer.

[0099] In one example, the server 210 may provide a configuration interface, and the configuration interface is used to configure the module 210. The user may send a configuration operation to the server through the client, and the configuration operation is used to configure the module 210. The server 210 may configure the module 210 based on the configuration operation, and send the configured module 210 to the user's local computer. For example, the configuration operation is to use the interpreter as the module 210, and the server 210 sends the interpreter as the module 210 to the user's local computer based on the configuration operation. For another example, the configuration operation is to use the first network layer in the interpreter as the module 210, and the server 210 sends the first network layer as the module 210 to the user's local computer based on the configuration operation. For another example, the configuration operation is to use the third network layer in the interpreter and the generator as the module 210, and the server 210 sends the interpreter and the third network layer as the module 210 to the user's local computer based on the configuration operation.

[0100] In some embodiments, the user can adjust the parameters of module 210. As described above, module 210 may include at least one network layer, wherein each network layer includes multiple neurons, and each neuron uses parameters to calculate data to obtain the processing result of the layer. In this way, the user adjusts the parameters so that the cloud or other parties cannot perceive the parameters of module 210, thereby making it more difficult for the cloud or other parties to reverse the content generation instructions based on the intermediate results sent by the client.

[0101] The user may issue a parameter adjustment operation, and the client 110 may receive the parameter adjustment operation and adjust the parameters of the module 210 based on the parameter adjustment operation.

[0102] In one example, the user can open the file of the module 210 through the client 110 using an application such as Notepad and directly modify the parameters of the module 210. That is, the parameter adjustment operation issued by the user includes an operation of opening the file of the module 210 and an operation of modifying the parameters. The client 110 can respond to the operation of opening the file of the module 210, open the file of the module 210, and respond to the operation of modifying the parameters to modify the parameters of the module 210.

[0103] In one example, the model 200 can be retrained locally on the user, and the parameters of the module 210 can be changed through retraining. Specifically, the server 210 can send the model 200 to the user locally. At the user's local, the model 200 can continue to be iteratively trained, for example, iterated once or multiple times to modify the parameters of the module 210. That is, the parameter adjustment operation issued by the user includes an operation to trigger retraining and an operation to terminate retraining. The client 110 can respond to the operation to trigger retraining and retrain the model 200, and respond to the operation to terminate retraining and terminate the training of the model 200.

[0104] In summary, in the content generation method provided in the embodiment of the present application, a part of the model for generating the target content is deployed locally on the user, and another part is deployed in the cloud. The part deployed locally on the user can perform data desensitization and other processing on the content generation instruction to obtain an intermediate result that is difficult to be used to infer the content generation instruction. Then, the intermediate result is sent to the cloud, so that the cloud generates the target content based on the intermediate result, thereby avoiding the leakage of sensitive information in the content generation instruction without requiring the user to run the entire model locally.

[0105] Based on the above description, the present application embodiment provides a content generation system 900. The system 900 is used to generate content using a model. Fig. 9As shown, the system 900 includes a client 910 and a server 920, the model includes a first module and a second module, the client 910 and the first module are deployed locally, the server 920 and the second module are deployed in the cloud, and the output of the first module is the input of the second module.

[0106] The client 910 is used to receive a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention;

[0107] The client 910 is used to input the content generation instruction into the first module, so that the first module outputs the intermediate result of the model generating the target content;

[0108] The server 920 is used to receive the intermediate result sent by the client 910, and input the intermediate result to the second module, so that the second module outputs the target content;

[0109] The server 920 is used to send the target content to the client 920 .

[0110] In some embodiments, the model is a generative model including multiple network layers, and the multiple network layers are arranged sequentially in the data transmission direction of the generative model, wherein the first module includes at least one network layer among the multiple network layers, and the second module includes network layers among the multiple network layers except the at least one network layer.

[0111] In some embodiments, the content generation instruction includes the user's intention, the model includes an interpreter and a generator, the interpreter is used to extract the characteristics of the user's intention to obtain an encoding vector, and the generator is used to generate the target content based on the encoding vector; wherein, the first module includes the interpreter, and the second module includes the generator; or, the first module includes the first network layer in the interpreter, and the second module includes the second network layer in the interpreter and the generator; the output of the first network layer is the input of the second network layer; or, the first module includes the interpreter and the third network layer in the generator, and the second module includes the fourth network layer in the generator; the output of the third network layer is the input of the fourth network layer.

[0112] In some embodiments, the model is a generative model, and the server 920 is further used to: train the model; after the model training is completed, send the first module to the user locally.

[0113] In some embodiments, the client 910 is further used to: receive a parameter adjustment operation issued by a user, and adjust the parameters of the first module based on the parameter adjustment operation.

[0114] In some embodiments, the first module includes a sensitive information filter set by the user, and the filter is used to intercept sensitive information in the content generation instruction to obtain the intermediate result.

[0115] The client 910 and the server 920 can be implemented by software or hardware. As an example, the implementation of the client 910 is described below. Similarly, the implementation of the server 920 can refer to the implementation of the client 910.

[0116] As an example of a software functional unit, the client 910 may include code running on a computing instance. Among them, the computing instance may be at least one of a physical host (computing device), a virtual machine, a container and other computing devices. Furthermore, the above-mentioned computing device may be one or more. For example, the client 910 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the application can be distributed in the same region or in different regions. The multiple hosts / virtual machines / containers used to run the code can be distributed in the same availability zone (AZ) or in different AZs, each AZ including a data center or multiple data centers with close geographical locations. Among them, usually a region can include multiple AZs.

[0117] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway must be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0118] As an example of a hardware functional unit, the client 910 may include at least one computing device, such as a server, a terminal, etc. Alternatively, the client 910 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0119] The multiple computing devices included in the client 910 can be distributed in the same region or in different regions. The multiple computing devices included in the client 910 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the client 910 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0120] Based on the above description, the embodiment of the present application provides a method for generating content using a model. The method can be applied to a client deployed locally on a user, wherein the content generation system where the client is located also includes a server deployed in the cloud, and the model includes a first module and a second module, wherein the first module is deployed locally on the user, and the second module is deployed in the cloud, and the output of the first module is the input of the second module. Fig.10 As shown, the method comprises the following steps:

[0121] Step 1001, the client receives a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention;

[0122] Step 1002: the client inputs the content generation instruction into the first module, so that the first module outputs the intermediate result of the model generating the target content;

[0123] Step 1003, the client sends the intermediate result to the server; wherein the server is used to input the intermediate result to the second module, so that the second module outputs the target content;

[0124] Step 1004: the client receives the target content sent by the server.

[0125] In some embodiments, the method further includes: the client receiving a parameter adjustment operation issued by a user, and adjusting the parameters of the first module based on the parameter adjustment operation.

[0126] See also Fig.11 The embodiment of the present application also provides a client 1100 for generating content using a model. The client 1100 is deployed locally on the user, and the content generation system where the client 1100 is located also includes a server deployed in the cloud. The model includes a first module and a second module, the first module is deployed locally on the user, the second module is deployed in the cloud, and the output of the first module is the input of the second module. The client 1100 includes:

[0127] A receiving module 1110 is used to receive a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention;

[0128] An input module 1120, configured to input the content generation instruction into the first module, so that the first module outputs an intermediate result of the model generating the target content;

[0129] The communication module 1130 is used to send the intermediate result to the server; wherein the server is used to input the intermediate result to the second module so that the second module outputs the target content;

[0130] The communication module 1130 is further configured to receive the target content sent by the server.

[0131] Among them, the receiving module 1110, the input module 1120 and the communication module 1130 can all be implemented by software, or can be implemented by hardware. Exemplarily, the implementation of the receiving module 1110 is introduced below by taking the receiving module 1110 as an example. Similarly, the implementation of the input module 1120 and the communication module 1130 can refer to the implementation of the receiving module 1110.

[0132] As an example of a software functional unit, the receiving module 1110 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the receiving module 1110 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code can be distributed in the same region or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code can be distributed in the same AZ or in different AZs, and each AZ includes a data center or multiple data centers with close geographical locations. Among them, usually a region can include multiple AZs.

[0133] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same VPC or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0134] As an example of a hardware functional unit, the receiving module 1110 may include at least one computing device, such as a server, etc. Alternatively, the receiving module 1110 may also be a device implemented using an ASIC or a programmable logic device PLD, etc. The PLD may be implemented using a CPLD, an FPGA, a GAL, or any combination thereof.

[0135] The multiple computing devices included in the receiving module 1110 can be distributed in the same region or in different regions. The multiple computing devices included in the receiving module 1110 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the receiving module 1110 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0136] It should be noted that, in other embodiments, the receiving module 1110 may be used to execute Fig.10 In any step of the method shown, the input module 1120 can be used to perform Fig.10 In any step of the method shown, the communication module 1130 can be used to perform Fig.10The steps implemented by the receiving module 1110, the input module 1120 and the communication module 1130 can be specified as needed, and the receiving module 1110, the input module 1120 and the communication module 1130 can be respectively implemented Fig.10 The different steps in the method shown are used to implement the entire functionality of the client 1100 .

[0137] The present application also provides a computing device 1200. Fig.12 As shown, the computing device 1200 includes: a bus 1202, a processor 1204, a memory 1206, and a communication interface 1208. The processor 1204, the memory 1206, and the communication interface 1208 communicate through the bus 1202. The computing device 1200 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 1200.

[0138] The bus 1202 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.12 The bus 1202 may include a path for transmitting information between various components of the computing device 1200 (eg, the memory 1206, the processor 1204, and the communication interface 1208).

[0139] The processor 1204 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0140] The memory 1206 may include a volatile memory, such as a random access memory (RAM). The memory 1206 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0141] The memory 1206 stores executable program codes, and the processor 1204 executes the executable program codes to respectively implement the functions of the aforementioned receiving module 1110, the input module 1120, and the communication module 1130, thereby implementing Fig.10 That is, the memory 1206 stores the method for executing Fig.10 Instructions for the method shown.

[0142] The communication interface 1208 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 1200 and other devices or communication networks.

[0143] The embodiment of the present application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smart phone.

[0144] like Fig.13 As shown, the computing device cluster includes at least one computing device 1200. The memory 1206 in one or more computing devices 1200 in the computing device cluster may store the same Fig.10 Instructions for the method shown.

[0145] In some possible implementations, the memory 1206 of one or more computing devices 1200 in the computing device cluster may also store a program for executing Fig.10 In other words, the combination of one or more computing devices 1200 can jointly execute instructions for executing Fig.10 Instructions for the method shown.

[0146] It should be noted that the memory 1206 in different computing devices 1200 in the computing device cluster may store different instructions, which are respectively used to execute part of the functions of the client 1100. That is, the instructions stored in the memory 1206 in different computing devices 1200 may implement the functions of one or more modules among the receiving module 1110, the input module 1120 and the communication module 1130.

[0147] In some possible implementations, one or more computing devices in the computing device cluster may be connected via a network, which may be a wide area network or a local area network. Fig.14 A possible implementation is shown. Fig.14As shown, two computing devices 1200A and 1200B are connected via a network. Specifically, the network is connected via a communication interface in each computing device. In this type of possible implementation, the memory 1206 in the computing device 1200A stores instructions for executing the functions of the receiving module 1110. At the same time, the memory 1206 in the computing device 1200B stores instructions for executing the functions of the input module 1120 and the communication module 1130.

[0148] It should be understood that Fig.14 The functions of the computing device 1200A shown in FIG. 1200A may also be completed by multiple computing devices 1200. Similarly, the functions of the computing device 1200B may also be completed by multiple computing devices 1200.

[0149] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similar to that of Fig.13 and Fig.14 The difference is that the memory 1206 in one or more computing devices 1200 in the computing device cluster may store the same memory for executing Fig.10 Instructions for the method shown.

[0150] In some possible implementations, the memory 1206 of one or more computing devices 1200 in the computing device cluster may also store a program for executing Fig.10 In other words, the combination of one or more computing devices 1200 can jointly execute instructions for executing Fig.10 Instructions for the method shown.

[0151] The present application also provides a computer program product including instructions. The computer program product may be software or a program product including instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes Fig.10 The method shown.

[0152] The present application also provides a computer-readable storage medium. The computer-readable storage medium may be any available medium that can be stored by a computing device or a host migration device such as a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute Fig.10 The method shown.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating content using a model, characterized in that: Applied to a content generation system, the system includes a client and a server, the model includes a first module and a second module, the client and the first module are deployed locally on the user, the server and the second module are deployed in the cloud, and the output of the first module is the input of the second module; the method includes: The client receives a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; The client inputs the content generation instruction into the first module, so that the first module outputs an intermediate result of the model generating the target content; The server receives the intermediate result sent by the client, and inputs the intermediate result to the second module, so that the second module outputs the target content; The server sends the target content to the client.

2. The method according to claim 1, characterized in that The model is a generative model including multiple network layers, and the multiple network layers are arranged sequentially in the data transmission direction of the generative model, wherein the first module includes at least one network layer among the multiple network layers, and the second module includes network layers among the multiple network layers except the at least one network layer.

3. The method according to claim 1 or 2, characterized in that: The content generation instruction includes the user's intention, the model includes an interpreter and a generator, the interpreter is used to extract the characteristics of the user's intention to obtain a coding vector, and the generator is used to generate the target content based on the coding vector; wherein, The first module includes the interpreter, and the second module includes the generator; or, The first module includes the first network layer in the interpreter, and the second module includes the second network layer in the interpreter and the generator; the output of the first network layer is the input of the second network layer; or, The first module includes the interpreter and the third network layer in the generator, and the second module includes the fourth network layer in the generator; the output of the third network layer is the input of the fourth network layer.

4. The method according to any one of claims 1 to 3, characterized in that The model is a generative model, and the method further comprises: The server trains the model; After the model training is completed, the server sends the first module to the user locally.

5. The method according to any one of claims 1 to 4, characterized in that The method further includes: the client receiving a parameter adjustment operation issued by a user, and adjusting the parameters of the first module based on the parameter adjustment operation.

6. The method according to claim 1, characterized in that The first module includes a sensitive information filter set by the user, and the filter is used to intercept sensitive information in the content generation instruction to obtain the intermediate result.

7. A content generation system, characterized in that: The system is used to generate content using a model, wherein the system includes a client and a server, the model includes a first module and a second module, the client and the first module are deployed locally on the user, the server and the second module are deployed in the cloud, and the output of the first module is the input of the second module; The client is used to receive a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; The client is used to input the content generation instruction into the first module, so that the first module outputs the intermediate result of the model generating the target content; The server is used to receive the intermediate result sent by the client, and input the intermediate result to the second module, so that the second module outputs the target content; The server is used to send the target content to the client.

8. The system according to claim 7, characterized in that The model is a generative model including multiple network layers, and the multiple network layers are arranged sequentially in the data transmission direction of the generative model, wherein the first module includes at least one network layer among the multiple network layers, and the second module includes network layers among the multiple network layers except the at least one network layer.

9. The system according to claim 7 or 8, characterized in that: The content generation instruction includes the user's intention, the model includes an interpreter and a generator, the interpreter is used to extract the characteristics of the user's intention to obtain a coding vector, and the generator is used to generate the target content based on the coding vector; wherein, The first module includes the interpreter, and the second module includes the generator; or, The first module includes the first network layer in the interpreter, and the second module includes the second network layer in the interpreter and the generator; the output of the first network layer is the input of the second network layer; or, The first module includes the interpreter and the third network layer in the generator, and the second module includes the fourth network layer in the generator; the output of the third network layer is the input of the fourth network layer.

10. The system according to any one of claims 7 to 9, characterized in that: The model is a generative model, and the server is further used for: training the model; After the model training is completed, the first module is sent to the user locally.

11. The system according to any one of claims 7 to 10, characterized in that: The client is further used to: receive a parameter adjustment operation issued by a user, and adjust the parameters of the first module based on the parameter adjustment operation.

12. The system according to claim 7, characterized in that The first module includes a sensitive information filter set by the user, and the filter is used to intercept sensitive information in the content generation instruction to obtain the intermediate result.

13. A method for generating content using a model, characterized in that: The method is applied to a client deployed locally on a user, wherein the content generation system where the client is located further includes a server deployed in the cloud, the model includes a first module and a second module, the first module is deployed locally on the user, the second module is deployed in the cloud, and the output of the first module is the input of the second module; the method includes: The client receives a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; The client inputs the content generation instruction into the first module, so that the first module outputs an intermediate result of the model generating the target content; The client sends the intermediate result to the server; wherein the server is used to input the intermediate result to the second module, so that the second module outputs the target content; The client receives the target content sent by the server.

14. The method according to claim 13, characterized in that The method further includes: the client receiving a parameter adjustment operation issued by a user, and adjusting the parameters of the first module based on the parameter adjustment operation.

15. A client for generating content using a model, characterized in that: The client is deployed locally on the user, and the content generation system where the client is located also includes a server deployed on the cloud. The model includes a first module and a second module. The first module is deployed locally on the user, and the second module is deployed on the cloud. The output of the first module is the input of the second module. The client includes: A receiving module, used to receive a content generation instruction issued by a user, wherein the content generation instruction is used to instruct the model to generate target content according to the user's intention; An input module, used for inputting the content generation instruction into the first module, so that the first module outputs an intermediate result of the model generating the target content; A communication module, used to send the intermediate result to the server; wherein the server is used to input the intermediate result to the second module, so that the second module outputs the target content; The communication module is also used to receive the target content sent by the server.

16. The client according to claim 15, characterized in that: The receiving module is further used to: receive a parameter adjustment operation issued by a user, and adjust the parameters of the first module based on the parameter adjustment operation.

17. A computing device cluster, characterized in that: comprising at least one computing device, each computing device comprising a processor and a memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster executes the method according to claim 13 or 14.

18. A computer program product comprising instructions, characterized in that When the instructions are executed by a computing device cluster, the computing device cluster executes the method according to claim 13 or 14.

19. A computer-readable storage medium, characterized in that: The method comprises computer program instructions, and when the computer program instructions are executed by a computing device cluster, the computing device cluster performs the method according to claim 13 or 14.

Images