Content generation and model updating method, device and equipment based on large model

Abstract

The invention provides a content generation and model updating method, device and equipment based on a large model, relates to the technical field of artificial intelligence, in particular to the technical fields of computer vision, deep learning, large models and the like, and can be applied to AIGC content generation and other scenes based on artificial intelligence. The model updating method comprises the steps of obtaining real preference data of a target object for generated content, wherein the generated content is obtained by an initial large model; scoring non-feedback contents except feedback contents corresponding to the real preference data by adopting a scoring model so as to obtain automatic preference data; obtaining preference sample data based on the real preference data and the automatic preference data; and based on the preference sample data, updating the initial large model to obtain a target large model.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, especially to technical fields such as computer vision, deep learning, and large models, and can be applied to scenarios such as AIGC content generation based on artificial intelligence, and specifically to a content generation and model update method, device, equipment, medium, and product based on large models. Background Art

[0002] Artificial Intelligence Generated Content (AIGC) refers to a technology that generates relevant content with appropriate generalization capabilities through learning and identifying existing data based on artificial intelligence technologies such as generative adversarial networks and large pre-trained models. Summary of the invention

[0003] The present disclosure provides a method, apparatus, device, medium and product for content generation and model updating based on a large model.

[0004] According to one aspect of the present disclosure, a model updating method is provided, including: obtaining real preference data of a target object for generated content, wherein the generated content is obtained by an initial large model; using a scoring model to score non-feedback content other than feedback content corresponding to the real preference data to obtain automatic preference data; obtaining preference sample data based on the real preference data and the automatic preference data; and updating the initial large model based on the preference sample data to obtain a target large model.

[0005] According to another aspect of the present disclosure, a content generation method based on a large model is provided, comprising: obtaining prompt information; using a target large model to generate content based on the prompt information; wherein the target large model is obtained using any of the methods described above.

[0006] According to another aspect of the present disclosure, a model updating device is provided, including: a first acquisition module, used to acquire real preference data of a target object for generated content, wherein the generated content is acquired by an initial large model; a second acquisition module, used to score non-feedback content other than feedback content corresponding to the real preference data using a scoring model, so as to acquire automatic preference data; a third acquisition module, used to acquire preference sample data based on the real preference data and the automatic preference data; and an updating module, used to update the initial large model based on the preference sample data, so as to obtain a target large model.

[0007] According to another aspect of the present disclosure, a content generation device based on a large model is provided, comprising: an acquisition module for acquiring prompt information; a generation module for using a target large model to generate content based on the prompt information; wherein the target large model is obtained using any of the methods described above.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any method as described in any of the above aspects.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the methods according to any one of the above aspects.

[0010] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the methods described in any one of the above aspects.

[0011] According to the embodiments of the present disclosure, a content generation model capable of dynamically tracking preferences can be obtained.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0015] Figure 2 is a schematic diagram of an implementation system for implementing an embodiment of the present disclosure;

[0016] Figure 3 is a schematic diagram of the overall architecture for implementing the embodiments of the present disclosure;

[0017] Figure 4 is a schematic diagram according to a second embodiment of the present disclosure;

[0018] Figure 5 is a schematic diagram according to a third embodiment of the present disclosure;

[0019] Figure 6 is a comparative schematic diagram of content generation effects provided according to an embodiment of the present disclosure;

[0020] Figure 7 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0021] Figure 8 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0022] Fig. 9 It is a schematic diagram of an electronic device used to implement the content generation model updating method or the content generation method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0023] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0024] AIGC technology can be applied in text-generated graph scenarios, that is, users input text into the text-generated graph model, and the text-generated graph model generates an image based on the text.

[0025] In order to improve user experience, it is necessary to generate images that meet user preferences.

[0026] In order to generate images that meet user preferences, expert annotation or crowdsourcing annotation is usually used to obtain preference data, and the preference data is used to train the text graph model.

[0027] However, since user preferences may change dynamically over time, the above method is difficult to cope with such dynamic changes, and there may be deviations between expert annotations or crowdsourcing annotations and user preferences.

[0028] In order to improve the effect of the content generation model, the present disclosure provides the following embodiments.

[0029] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure. This embodiment provides a model updating method, such as Figure 1 As shown, the method includes:

[0030] 101. Obtain real preference data of a target object for generated content, where the generated content is obtained by an initial large model.

[0031] 102. Use a scoring model to score non-feedback content other than the feedback content corresponding to the real preference data to obtain automatic preference data.

[0032] 103. Obtain preference sample data based on the real preference data and the automatic preference data.

[0033] 104. Based on the preference sample data, update the initial large model to obtain a target large model.

[0034] The target object refers to the object using the content generation model, such as a user.

[0035] The large model refers to the Large Language Model (LLM). LLM is a hot technology in the AI ​​field in recent years. LLM is a natural language processing model based on deep learning. It has a huge number of parameters and a complex structure, which enables it to process and understand a large amount of natural language data. Through the process of pre-training and fine-tuning, the LLM model can perform well in a variety of natural language processing tasks, including but not limited to text generation, language understanding, machine translation, etc.

[0036] Taking the text-image scenario as an example, the large model can be called a text-image model. In this scenario, users can input text into the text-image model, and the output of the text-image model is an image.

[0037] The above takes the text-to-image scenario as an example, and can also be applied to other scenarios, such as text-to-text (both input and output are text), text-to-video (input is text, output is video), etc.

[0038] For the sake of distinction, the content generation model before updating (such as the text-graph model) is called the initial large model, and the content generation model after updating is called the target large model.

[0039] After the initial large model obtains the generated content, the target object (such as the user) can give feedback on the generated content, such as likes or dislikes.

[0040] The real preference data is used to characterize the preferences of the target object. Specifically, it can be obtained based on the feedback data of the target object. For example, if a user likes an image, it indicates that the image meets the user's preferences. On the contrary, if a user dislikes an image, it indicates that the image does not meet the user's preferences.

[0041] Scoring models for obtaining automatic preference data.

[0042] Among them, users can give feedback on the images generated by the Wensheng map. The preference data obtained based on the feedback data can be used as real preference data, and the preference data obtained based on non-feedback data can be automatic preference data. Both the real preference data and the automatic preference data are used to represent user preferences.

[0043] Automatic preference data is data obtained based on a scoring model and used to characterize user preferences.

[0044] The preference sample data is sample data used to update the content generation model.

[0045] After obtaining the real preference data and automatic preference data of the target object, preference sample data can be obtained based on these real preference data and automatic preference data. For example, for an image that a user likes, the image is used as a positive sample, and for an image that a user dislikes, the image is used as a negative sample; samples that represent positive preferences in the automatic preference data are used as positive samples, and samples that represent negative preferences in the automatic preference data are used as negative samples; these positive samples and negative samples can constitute preference sample data.

[0046] After obtaining the preference sample data, the preference sample data can be used to update the initial large model to obtain the target large model.

[0047] In this embodiment, preference sample data is obtained based on the real preference data of the target object. Compared with expert labeling or crowdsourcing labeling, it can reflect the real preference needs of the target object and obtain a content generation model that is more accurate and more in line with the target object's preferences. The initial large model is updated based on the preference sample data to improve the dynamic adaptability of the model, thereby obtaining a content generation model that can dynamically track the target object's preferences. In addition, by obtaining automatic preference data, the sample data can be expanded to improve the model performance.

[0048] In order to better understand the present disclosure, the application scenarios involved in the present disclosure are described as follows:

[0049] Figure 2 It is a schematic diagram of an implementation system for implementing the embodiments of the present disclosure.

[0050] like Figure 2 As shown, the implementation system includes: a user terminal 201 and a server 202.

[0051] The user terminal 201 may include: a personal computer (PC), a laptop, a mobile phone or other mobile devices, and the server 202 may be a local server or a cloud server, and may be a single server or a cluster server.

[0052] A client is deployed on the user terminal 201 , and a server is deployed on the server 202 .

[0053] The server can generate content based on the large model (LLM).

[0054] For large models, the large model performs related operations according to prompt information.

[0055] To this end, the user can send a prompt message (prompt) to the server through the client, and the server calls the large model (LLM) to generate content based on the prompt message.

[0056] like Figure 2 As shown, taking the text-image scenario as an example, the prompt information is text, LLM generates an image based on the input text, and the output is an image.

[0057] In the related art, LLM is trained based on expert labeled data or crowdsourced labeled data, and is manually updated using these labeled data when the update conditions are met, which makes it difficult to dynamically track user preferences.

[0058] In order to dynamically track user preferences, in the embodiments of the present disclosure, the content generation model may be dynamically updated based on the user preferences.

[0059] Figure 3 It is a schematic diagram of the overall architecture for implementing the embodiments of the present disclosure.

[0060] like Figure 3 As shown in FIG, taking the text-generated graph scenario as an example, the corresponding content generation model is called the text-generated graph model, and the target object is the user as an example.

[0061] Generally speaking, it can be divided into online reasoning and model optimization. The document graph model used in online reasoning is called online document graph model 301. The document graph model trained in model optimization is called offline document graph model 302.

[0062] During online reasoning, the user can input prompt information, such as text, and the online text graph model 301 generates an image based on the text. The user provides feedback related to the image generated by the online text graph model 301, such as Figure 3 In the example, a check mark indicates a thumbs-up, i.e., the user likes the corresponding image, and a wrong mark indicates a thumbs-down, i.e., the user does not like the corresponding image. These feedback data can be stored in the database as the user's real preference data. The image that the user likes is taken as a positive sample, and otherwise as a negative sample.

[0063] Since the amount of user feedback data is relatively small, in order to expand the number of samples, a scoring model may be used to score the non-feedback content, and automatic preference data may be obtained based on the scoring results, and the automatic preference data may be stored in the database 303 .

[0064] Scoring model 304 is used to obtain automatic preference data.

[0065] Among them, the user can provide feedback on the image generated by the online text graph model 301, the preference data obtained based on the feedback data can be used as real preference data, and the preference data obtained based on non-feedback data can be automatic preference data. Both the real preference data and the automatic preference data are used to represent user preferences.

[0066] The automatic preference data is data obtained according to the scoring model 304 and used to characterize the user's preference. The automatic preference data may also be stored in the database 303.

[0067] Furthermore, the scoring model 304 may be trained based on the user's actual preference data.

[0068] The input of the scoring model 304 is a sample pair of <prompt information, image>, and the output is an image score. Images with scores higher than a preset value are taken as positive samples, and images with scores lower than the preset value are taken as negative samples.

[0069] Furthermore, in order to improve timeliness, the scoring model can be updated regularly, that is, when a preset training cycle is reached, the scoring model is retrained using the real preference data to obtain an updated scoring model, and the updated scoring model is used to re-score, thereby updating the automatic preference data.

[0070] After obtaining the real preference data and the automatic preference data, they can be combined into preference sample data, and the preference sample data can be used to update the text graph model.

[0071] Furthermore, these preference data may be stored in a preference database, for example, the real preference data may be stored in a first storage space of the preference database, and the automatic preference data may be stored in a second storage space of the preference database.

[0072] In this way, data management can be facilitated by storing different types of data separately.

[0073] For example, real preference data can be obtained from the first storage space, and the scoring model can be trained using the real preference data. After automatic preference data is obtained based on the scoring model, the automatic preference data is stored in the second storage space.

[0074] When the model is updated using the preference sample data, the offline document graph model 302 may be trained using the preference sample data, and when a preset update cycle is reached, the offline document graph model 302 is used to replace parameters of the online document graph model 301 .

[0075] The above-mentioned image generation-user feedback-text graph model update can be continuously executed in a cycle to realize automatic update of the text graph model and dynamically track user preferences.

[0076] In combination with the above application scenarios, the present disclosure also provides the following embodiments.

[0077] Figure 4 is a schematic diagram according to the second embodiment of the present disclosure, which provides a model updating method. Figure 4 As shown, the method includes:

[0078] 401. Use the initial large model to obtain generated content.

[0079] 402. Obtain real preference data of the target object for the generated content.

[0080] For example, refer to Figure 3 , Taking the target object as a user as an example, the user can give feedback operations such as likes or dislikes to the images obtained by the text graph model. Based on these preference-related feedback data, the real preference data is obtained and stored in the database.

[0081] 403. Using a scoring model, score the non-feedback content other than the feedback content corresponding to the real preference data to obtain automatic preference data.

[0082] 404. Obtain preference sample data based on the real preference data and the automatic preference data.

[0083] For example, refer to Figure 3 , the prompt information in the non-feedback content and its corresponding image are input into the scoring model, and the output is the score. Automatic preference data is obtained based on the score. If the score is higher than the preset value, it indicates that the corresponding image is a positive sample, otherwise it is a negative sample.

[0084] The preference sample data includes real preference data and automatic preference data. Specifically, images that users like are taken as positive samples, otherwise they are taken as negative samples.

[0085] In this embodiment, automatic preference data is obtained based on the scoring model, and preference sample data is obtained based on the real preference data and the automatic preference data. This can expand the amount of preference sample data, improve the performance of the content generation model, and further improve the content generation effect.

[0086] In some embodiments, the scoring model can be trained based on the real preference data.

[0087] The real preference data may be used as training samples of the scoring model, and the scoring model may be trained using the real preference data.

[0088] In this embodiment, the scoring model is trained based on the real preference data, so that the automatic preference data obtained by the scoring model can be more in line with the preferences of the target object, thereby improving the accuracy of preference tracking.

[0089] In some embodiments, in response to reaching a preset training period, the scoring model can be trained based on the real preference data.

[0090] In this way, through the preset training cycle, the scoring model can be updated regularly, so that the scoring model can be updated more timely, the timeliness of the scoring model can be improved, and the accuracy and dynamism of preference tracking can be improved.

[0091] 405. In response to reaching a preset update cycle, the initial large model is updated based on the preference sample data to obtain a target large model.

[0092] In this way, the content generation model can be updated regularly through a preset update cycle, so that content that meets the preferences of the target objects can be obtained in a more dynamic and timely manner.

[0093] In some embodiments, an offline content generation model can be trained based on the preference sample data; in response to reaching a preset update cycle, the model parameters of the initial large model are replaced with the model parameters of the offline content generation model to obtain the target large model.

[0094] In this embodiment, an offline content generation model is obtained by offline training, which can avoid interference with the online reasoning process, improve training efficiency and ensure data security. By regularly replacing model parameters, the content generation model can be regularly updated to obtain a more dynamic content generation model that is more in line with the preferences of the target object, thereby obtaining generated content that dynamically tracks preferences.

[0095] In some embodiments, the preference sample data includes: prompt information and image samples, and the image samples include positive samples and negative samples, the positive samples are used to characterize the positive preference of the target object, and the negative samples are used to characterize the negative preference of the target object;

[0096] The updating of the initial large model based on the preference sample data to obtain a target large model includes:

[0097] Inputting the prompt information into the initial large model to obtain a predicted image;

[0098] Constructing a loss function based on the predicted image and the image sample;

[0099] Based on the loss function, the model parameters of the initial large model are adjusted to obtain the target large model.

[0100] In this embodiment, a loss function is constructed based on the predicted image and the image samples, and then the model parameters are adjusted, without paying attention to the internal structure of the model, so that the model can be updated more simply and efficiently.

[0101] Figure 5 is a schematic diagram according to the third embodiment of the present disclosure, which provides a content generation method based on a large model, such as Figure 5 As shown, the method includes:

[0102] 501. Get prompt information.

[0103] 502. Use the target macro model to generate content based on the prompt information.

[0104] Wherein, the target large model is obtained by using the updating method shown in any of the above embodiments.

[0105] Figure 6 It is a comparative schematic diagram of the content generation effect provided according to the embodiment of the present disclosure.

[0106] like Figure 6 As shown in FIG. 1 , assuming that the prompt information is “the picture shows a painting of swan feathers decorated with leaves. The snow-white swan feathers contrast sharply with the emerald green leaves, and the leaves are cleverly inlaid between the swan feathers, adding a touch of natural breath. The swan feathers are distinct, revealing a kind of nobility and elegance. The whole painting is bright in color, beautifully composed, and the background is in light tones, which further highlights the decorative effect of the swan feathers and leaves”, based on the prompt information, the above target large model can be used to obtain Figure 6 The image shown on the left is obtained by using the model of related technology Figure 6 Image shown on the right.

[0107] Based on this comparison chart, we can see that by tracking user preferences, Figure 6 The image shown on the left has a stronger graphic relevance to the text (prompt information) and a better image quality.

[0108] In this embodiment, since the target macro model can dynamically track the preferences of the target object, content generation based on the target macro model can obtain generated content that dynamically tracks the preferences of the target object. When the target object is a user, content that is more in line with the user's preferences can be obtained, thereby improving the user experience.

[0109] Figure 7 is a schematic diagram according to a fourth embodiment of the present disclosure. This embodiment provides a model updating device, such as Figure 7 As shown, the device 700 includes: a first acquisition module 701, a second acquisition module 702, a third acquisition module 703 and an update module 704.

[0110] The first acquisition module 701 is used to obtain the real preference data of the target object for the generated content, and the generated content is obtained by the initial large model; the second acquisition module 702 is used to use a scoring model to score the non-feedback content other than the feedback content corresponding to the real preference data to obtain automatic preference data; the third acquisition module 703 is used to obtain preference sample data based on the real preference data and the automatic preference data; the update module 704 is used to update the initial large model based on the preference sample data to obtain the target large model.

[0111] In this embodiment, preference sample data is obtained based on the real preference data of the target object. Compared with expert labeling or crowdsourcing labeling, it can reflect the real preference needs of the target object and obtain a content generation model that is more accurate and more in line with the target object's preferences. The initial large model is updated based on the preference sample data to improve the dynamic adaptability of the model, thereby obtaining a content generation model that can dynamically track the target object's preferences. In addition, by obtaining automatic preference data, the sample data can be expanded to improve the model performance.

[0112] In some embodiments, the apparatus 700 further includes:

[0113] A training module is used to train the scoring model based on the real preference data.

[0114] In this embodiment, the scoring model is trained based on the real preference data, so that the automatic preference data obtained by the scoring model can be more in line with the preferences of the target object, thereby improving the accuracy of preference tracking.

[0115] In some embodiments, the training module is further used to:

[0116] In response to reaching a preset training period, the scoring model is trained based on the real preference data.

[0117] In this embodiment, the scoring model can be updated regularly through a preset training cycle, so that the scoring model can be updated more timely, thereby improving the timeliness of the scoring model and improving the accuracy and dynamics of preference tracking.

[0118] In some embodiments, the apparatus 700 further includes:

[0119] The storage module is used to store the real preference data in a first storage space of a preference database, and to store the automatic preference data in a second storage space of the preference database.

[0120] In this embodiment, data management can be facilitated by storing different types of data separately.

[0121] In some embodiments, the preference sample data includes: prompt information and image samples, and the image samples include positive samples and negative samples, the positive samples are used to characterize the positive preference of the target object, and the negative samples are used to characterize the negative preference of the target object;

[0122] The updating module 703 is further used for:

[0123] Inputting the prompt information into the initial large model to obtain a predicted image;

[0124] Constructing a loss function based on the predicted image and the image sample;

[0125] Based on the loss function, the model parameters of the initial large model are adjusted to obtain the target large model.

[0126] In this embodiment, a loss function is constructed based on the predicted image and the image samples, and then the model parameters are adjusted, without paying attention to the internal structure of the model, so that the model can be updated more simply and efficiently.

[0127] In some embodiments, the updating module 703 is further used to:

[0128] In response to reaching a preset update cycle, the initial large model is updated based on the preference sample data to obtain a target large model.

[0129] In this embodiment, the content generation model can be updated regularly through a preset update cycle, so that content that meets the preferences of the target object can be obtained more dynamically and timely.

[0130] In some embodiments, the updating module 703 is further used to:

[0131] Based on the preference sample data, training an offline content generation model;

[0132] In response to reaching a preset update cycle, the model parameters of the initial large model are replaced with the model parameters of the offline content generation model to obtain the target large model.

[0133] In this embodiment, an offline content generation model is obtained by offline training, which can avoid interference with the online reasoning process, improve training efficiency and ensure data security. By regularly replacing model parameters, the content generation model can be regularly updated to obtain a more dynamic content generation model that is more in line with the preferences of the target object, thereby obtaining generated content that dynamically tracks preferences.

[0134] Figure 8 is a schematic diagram according to the fifth embodiment of the present disclosure, and this embodiment provides a content generation device based on a large model, such as Figure 8As shown, the device 800 includes: an acquisition module 801 and a generation module 802.

[0135] The acquisition module 801 is used to acquire prompt information; the generation module 802 is used to use the target macro model to generate content based on the prompt information;

[0136] Wherein, the target large model is obtained by using the updating method shown in any of the above embodiments.

[0137] In this embodiment, since the target macro model can dynamically track the preferences of the target object, content generation based on the target macro model can obtain generated content that dynamically tracks the preferences of the target object. When the target object is a user, content that is more in line with the user's preferences can be obtained, thereby improving the user experience.

[0138] It can be understood that in the embodiments of the present disclosure, the same or similar contents in different embodiments can be referenced to each other.

[0139] It can be understood that the “first”, “second”, etc. in the embodiments of the present disclosure are only used for distinction and do not indicate the degree of importance, time sequence, etc.

[0140] It is understandable that unless there is any special limitation on the sequence of steps in the process, it means that the timing relationship between these steps is not limited.

[0141] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0142] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0143] Fig. 9 A schematic block diagram of an example electronic device 900 that can be used to implement an embodiment of the present disclosure is shown. The electronic device 900 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0144] like Fig. 9As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 909 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the electronic device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 909 is also connected to the bus 904.

[0145] Multiple components in the electronic device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the electronic device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0146] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as a content generation model update method or a content generation method. For example, in some embodiments, the content generation model update method or the content generation method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the content generation model update method or the content generation method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute the content generation model updating method or the content generation method in any other appropriate manner (eg, by means of firmware).

[0147] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0148] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable task processing device, so that the program code, when executed by the processor or controller, implements the functions / operations specified in the flow chart and / or block diagram. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0149] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0150] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0151] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0152] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0153] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0154] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A model updating method, comprising: Acquire the target object's real preference data for the generated content, where the generated content is obtained by the initial large model; Using a scoring model, scoring the non-feedback content other than the feedback content corresponding to the real preference data to obtain automatic preference data; Based on the real preference data and the automatic preference data, obtaining preference sample data; Based on the preference sample data, the initial large model is updated to obtain a target large model.

2. The method according to claim 1, further comprising: The scoring model is trained based on the real preference data.

3. The method according to claim 2, wherein: The step of training the scoring model based on the real preference data comprises: In response to reaching a preset training period, the scoring model is trained based on the real preference data.

4. The method according to claim 1, further comprising: The real preference data is stored in a first storage space of a preference database, and the automatic preference data is stored in a second storage space of the preference database.

5. The method according to claim 1, wherein: The preference sample data includes: prompt information and image samples, and the image samples include positive samples and negative samples, the positive samples are used to characterize the positive preference of the target object, and the negative samples are used to characterize the negative preference of the target object; The updating of the initial large model based on the preference sample data to obtain a target large model includes: Inputting the prompt information into the initial large model to obtain a predicted image; Constructing a loss function based on the predicted image and the image sample; Based on the loss function, the model parameters of the initial large model are adjusted to obtain the target large model.

6. The method according to claim 1, wherein: The updating of the initial large model based on the preference sample data to obtain a target large model includes: In response to reaching a preset update cycle, the initial large model is updated based on the preference sample data to obtain a target large model.

7. The method according to claim 6, wherein: In response to reaching a preset update cycle, the initial large model is updated based on the preference sample data to obtain a target large model, including: Based on the preference sample data, training an offline content generation model; In response to reaching a preset update cycle, the model parameters of the initial large model are replaced with the model parameters of the offline content generation model to obtain the target large model.

8. The method according to claim 1, wherein: The initial large model and the target large model are both Wensheng graph models.

9. A content generation method based on a large model, comprising: Get prompt information; Using a target macro model, generating content based on the prompt information; Wherein, the target large model is obtained by using the method described in any one of claims 1-8.

10. A content generating device, comprising: A first acquisition module is used to acquire the target object's real preference data for the generated content, where the generated content is obtained by the initial large model; A second acquisition module is used to score the non-feedback content other than the feedback content corresponding to the real preference data using a scoring model to obtain automatic preference data; A third acquisition module, configured to acquire preference sample data based on the real preference data and the automatic preference data; An updating module is used to update the initial large model based on the preference sample data to obtain a target large model.

11. The apparatus according to claim 10, further comprising: A training module is used to train the scoring model based on the real preference data.

12. The device according to claim 11, wherein The training module is further used to: In response to reaching a preset training period, the scoring model is trained based on the real preference data.

13. The apparatus according to claim 10, further comprising: The storage module is used to store the real preference data in a first storage space of a preference database, and to store the automatic preference data in a second storage space of the preference database.

14. The device according to claim 10, wherein: The preference sample data includes: prompt information and image samples, and the image samples include positive samples and negative samples, the positive samples are used to characterize the positive preference of the target object, and the negative samples are used to characterize the negative preference of the target object; The update module is further used for: Inputting the prompt information into the initial large model to obtain a predicted image; Constructing a loss function based on the predicted image and the image sample; Based on the loss function, the model parameters of the initial large model are adjusted to obtain the target large model.

15. The device according to claim 10, wherein: The update module is further used for: In response to reaching a preset update cycle, the initial large model is updated based on the preference sample data to obtain a target large model.

16. The device according to claim 15, wherein: The update module is further configured to: Based on the preference sample data, training an offline content generation model; In response to reaching a preset update cycle, the model parameters of the initial large model are replaced with the model parameters of the offline content generation model to obtain the target large model.

17. A content generation device based on a large model, comprising: An acquisition module is used to obtain prompt information; A generation module, used to generate content based on the prompt information using a target macro model; Wherein, the target large model is obtained by using the method described in any one of claims 1-8.

18. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

19. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-9.

20. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images