Text reasoning model training method, digital human text reasoning method and related equipment

Abstract

The present application discloses a text reasoning model training method, a digital human text reasoning method and related equipment, which belong to the field of artificial intelligence technology. Among them, the method includes: responding to a model training request initiated by a first terminal, obtaining a training file and a training instance identifier corresponding to the model training request; preprocessing the training file according to a pre-configured model training preprocessing tool to obtain model training data; assigning the model training data to a model learning service instance corresponding to the training instance identifier, obtaining a text reasoning model obtained by the model learning service instance training a pre-trained language model according to the model training data. In this way, the fine-tuning and training process of the pre-trained language model can be simplified, reducing the complexity of the operation, thereby making the model application easier to promote.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a text reasoning model training method, a digital human text reasoning method, and related equipment. Background Art

[0002] With the rise of Large Language Models (LLM), more and more large enterprises and individual developers have begun to participate in the fine-tuning and training of large models. However, fine-tuning large language models not only requires certain programming and language skills, but also requires developers to have strong operating system operation capabilities, especially in processing data, configuring the environment, and optimizing the training process. For ordinary users, it is still difficult to try to fine-tune and apply models before mastering these necessary skills. It can be seen that the existing technical complexity of fine-tuning and training of large models is relatively high, and it is difficult to promote it among the general population. Summary of the invention

[0003] The embodiments of the present application provide a text reasoning model training method, a digital human text reasoning method and related equipment to at least solve the problem that the existing large language model fine-tuning and training are complex and difficult to promote among the general population.

[0004] In order to solve the above technical problems, this application is implemented as follows: In a first aspect, an embodiment of the present application provides a text inference model training method, comprising: responding to a model training request initiated by a first terminal, obtaining a training file and a training instance identifier corresponding to the model training request; preprocessing the training file according to a pre-configured model training preprocessing tool to obtain model training data; and obtaining a text inference model obtained by training a pre-trained language model according to the model training data by the model learning service instance by assigning the model training data to a model learning service instance corresponding to the training instance identifier.

[0005] In the second aspect, an embodiment of the present application provides a digital human text reasoning method, including: responding to a business request initiated by a second terminal, obtaining text content and a training instance identifier corresponding to the business request; reasoning the text content through a text reasoning model corresponding to the training instance identifier to obtain a model reasoning result; wherein the text reasoning model is trained according to the method described in the first aspect; generating a target video stream based on the model reasoning result and a preset digital human image; and sending the target video stream to the second terminal.

[0006] In a third aspect, an embodiment of the present application provides a text reasoning model training device, comprising: a file acquisition module, used to respond to a model training request initiated by a first terminal, and obtain a training file and a training instance identifier corresponding to the model training request; a file processing module, used to preprocess the training file according to a pre-configured model training preprocessing tool to obtain model training data; a model training module, used to assign the model training data to a model learning service instance corresponding to the training instance identifier, and obtain a text reasoning model obtained by the model learning service instance training a pre-trained language model according to the model training data.

[0007] In the fourth aspect, an embodiment of the present application provides a digital human text reasoning device, including: a business request module, used to respond to a business request initiated by a second terminal, and obtain the text content and training instance identifier corresponding to the business request; a model reasoning module, used to reason about the text content through the text reasoning model corresponding to the training instance identifier, and obtain a model reasoning result; wherein the text reasoning model is trained according to the method described in the first aspect above; a video stream generation module, used to generate a target video stream according to the model reasoning result and a preset digital human image; a video stream sending module, used to send the target video stream to the second terminal.

[0008] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the method described in the first aspect above are implemented.

[0009] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect or the second aspect above are implemented.

[0010] In the seventh aspect, an embodiment of the present application provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of the method described in the first or second aspect above.

[0011] In an embodiment of the present application, in response to a model training request initiated by the first terminal, a training file and a training instance identifier corresponding to the model training request are obtained; the training file is preprocessed according to a pre-configured model training preprocessing tool to obtain model training data; and the model training data is assigned to a model learning service instance corresponding to the training instance identifier to obtain a text reasoning model obtained by training a pre-trained language model according to the model training data by the model learning service instance. In this way, by preprocessing the training file with a model training preprocessing tool and assigning the pre-processed model training data to a corresponding model learning service instance for model training, the fine-tuning and training process of the pre-trained language model can be simplified, and the operation complexity can be reduced, thereby making the model application easier to promote.

[0012] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0014] Figure 1 The following is an architecture diagram of a digital human processing system provided by an embodiment of the present application; Figure 2 A flow chart of a text reasoning model training method provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of the process of querying the model training progress provided by an embodiment of the present application is shown; Figure 4a and 4b A schematic diagram of the process of the LLM training method provided in an embodiment of the present application is shown; Figure 5 A schematic diagram of the process of the digital human text reasoning method provided in the embodiment of the present application is shown; Figure 6 A schematic diagram of a flow chart of a text reasoning processing method provided in an embodiment of the present application is shown; Figure 7 A schematic diagram of the process of the personalized reasoning method of a digital human provided in an embodiment of the present application is shown; Figure 8a and 8b A schematic diagram showing the process of the personalized image training method for digital humans provided in the embodiment of the present application is shown; Fig. 9 A schematic diagram of the process of a digital personalized training progress query method provided in an embodiment of the present application is shown; Fig.10A schematic diagram of the structure of a text reasoning model training device provided in an embodiment of the present application is shown; Fig.11 A schematic diagram showing the structure of a digital human text reasoning device provided in an embodiment of the present application is shown; Fig.12 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0015] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0016] In view of the high complexity of fine-tuning and training of existing large language models and the difficulty in promoting them among the general population, the present application embodiment provides a digital human processing system, such as Figure 1 As shown in , the digital human processing system 120 includes: load balancing 121, LLM training management module 122, model training preprocessing tool 123, seven-layer load 124, LLM reasoning service, LLM learning service, learning progress processor 125, database 126, authentication routing 127, learning result query service 128, message delivery platform 129, artificial intelligence agent assistant AI Agent 1210, message service 1211, knowledge base 1212, digital human terminal agent 1213, media stream load 1214, digital human image training management module 1215, digital human training preprocessing tool 1216, digital human learning service and digital human reasoning service. Among them, Message service 1211, used to connect to the second terminal 140 for uplink and downlink messages, and connect to AI Agent 1210 for message reasoning; AI Agent 1210 is used to connect to the knowledge base 1212 and the LLM reasoning service for reasoning, and decide whether to call the digital human's own task processing based on FunctionCall according to actual needs to complete work tasks other than chatting; The LLM training management module 122 runs in the CPU resource pool, encapsulates complex logic by calling the model training preprocessing tool 123, and calls the user-owned training instance in the GPU resource pool to complete the LLM training task submission; The model training preprocessing tool 123 is used to automatically convert the training files uploaded by ordinary users into model training data that can be recognized by LLM.

[0017] The digital human image training management module 1215 runs in the CPU resource pool, and completes the submission of the digital human image training task by calling the digital human training preprocessing tool 1216, encapsulating complex logic, and calling the user-attributed training instance in the GPU resource pool; A digital human training preprocessing tool 1216, used to automatically convert a video file corresponding to a digital human image uploaded by an ordinary user into image training data that can be recognized by a preset model; Knowledge base 1212, used to complete knowledge training and query result recall based on knowledge base architectures including but not limited to langchain; LLM learning service, deployed in the GPU resource pool, is used to complete the learning of LLM training data and output the training logs to the public persistent volume; LLM inference service, deployed in the GPU resource pool, is used to implement inference of user data; Digital human learning service, deployed in the GPU resource pool, is used to complete the learning of digital human personalized image training data and output the training log to the public persistent volume; Digital Human Reasoning Service, deployed in the GPU resource pool, is used to realize text content reasoning into the target video stream corresponding to the user image; A learning progress processor 125 is used to parse the training logs of the LLM learning service and the digital human learning service, and update the progress to the database 126; The learning result query service 128 is used to receive the training task progress query from the trainer.

[0018] Figure 2 The flowchart of the text reasoning model training method provided in the embodiment of the present application is shown. The execution subject of the method can be a server, wherein the server can be a background server of a certain business, or a background server of a certain platform or application (for example, a text reasoning model training platform, a digital human processing system, etc.). Figure 1 The LLM training management module 122 in the digital human processing system is used as an example for explanation. As shown in the figure, the method 200 may include the following steps: Step 201: respond to a model training request initiated by a first terminal, and obtain a training file and a training instance identifier corresponding to the model training request.

[0019] Among them, the first terminal can be a device such as a personal computer, or a mobile terminal device such as a mobile phone, a tablet computer, etc. The first terminal can be a terminal device used by a user; the training file can be a data set of conversation data, articles, behavioral norms or other types of text.

[0020] In a specific implementation, the first terminal 110 initiates a model training request to the digital human processing system 120. After receiving the front-end request, the load balancer 121 in the CPU resource pool performs load balancing, obtains the training file corresponding to the model training request, and determines that the training instance identifier tag parameter is i, and forwards tag_i and the training file to the LLM training management module 122. The LLM training management module 122 obtains the training file and the corresponding training instance identifier tag_i, and performs the subsequent asynchronous processing training process.

[0021] Step 202: pre-process the training file according to a pre-configured model training pre-processing tool to obtain model training data.

[0022] In a specific implementation, the LLM training management module 122 calls a pre-configured model training preprocessing tool 123 to convert the training file into model training data that can be understood and processed by the large model. For example, the model training preprocessing tool 123 performs data cleaning, word segmentation, labeling and other preprocessing on the training file to obtain the model training data.

[0023] Step 203: By assigning the model training data to the model learning service instance corresponding to the training instance identifier, the text inference model obtained by the model learning service instance through training the pre-trained language model according to the model training data.

[0024] In a specific implementation, the LLM training management module 122 obtains the training instance identifier tag_i corresponding to the training file, and distributes the model training data to the model learning service instance corresponding to tag_i through the seven-layer load 124 in the GPU resource pool, for example, the LLM learning service tag_i. The LLM learning service tag_i trains the pre-trained language model LLM according to the model training data to obtain a text reasoning model.

[0025] Here, the seven-layer load 124 performs load balancing at the application layer (seventh layer) and routes the model training data to the LLM learning service instance tag_i according to tag_i.

[0026] In this way, by preprocessing the training files through the model training preprocessing tool and assigning the preprocessed model training data to the corresponding model learning service instance for model training, the fine-tuning and training process of the pre-trained language model can be simplified, and the operation complexity can be reduced, thereby making the model application easier to promote. At the same time, by assigning model training data through training instance identification, multi-user isolated training can be achieved, ensuring that multiple users can perform model training at the same time, thereby improving the training efficiency of the text reasoning model.

[0027] In a possible implementation, in the above step 201, responding to the model training request initiated by the first terminal, obtaining the training file and the training instance identifier corresponding to the model training request includes: In response to a training request initiated by a first terminal, a first user identifier corresponding to the first terminal is obtained by performing user authentication on the first terminal; a training file corresponding to the training request is obtained, and a training instance identifier corresponding to the training file is determined according to the first user identifier.

[0028] In an embodiment of the present application, after obtaining the training request initiated by the first terminal 110, the load balancing 121 in the CPU resource pool can call the authentication routing 127 to perform user authentication on the first terminal 110, obtain the first user identifier corresponding to the first terminal, obtain the training file corresponding to the training request, and determine the training instance identifier tag_i corresponding to the first user identifier based on the pre-stored mapping relationship between the user identifier and the training instance identifier, and determine tag_i as the training instance identifier corresponding to the training file.

[0029] Optionally, if the user identification ID is not bound to a corresponding training instance identifier, a new training instance identifier is allocated to the user identification ID.

[0030] In a possible implementation, in the above step 203, allocating the model training data to the model learning service instance corresponding to the training instance identifier includes: Create a model training task and determine a first task identifier corresponding to the model training task; route the first task identifier and the model training data to a model learning service instance corresponding to the training instance identifier.

[0031] In a specific implementation, after acquiring the model training data and tag_i, the LLM training management module 122 creates a model training task, determines the first task identifier taskId corresponding to the model training task, and sends the model training data, taskId and tag_i to the seven-layer load 124. The seven-layer load 124 parses tag_i and routes it to the corresponding LLM learning service instance tag_i.

[0032] Wherein, after routing the first task identifier and the model training data to the model learning service instance corresponding to the training instance identifier, the method further includes: Obtain a first progress query request sent by the first terminal, wherein the first progress query request includes a first task identifier; obtain a first model training progress of the pre-trained language model according to the first task identifier; generate model training progress information corresponding to the training progress query request according to the first model training progress; and send the model training progress information to the first terminal.

[0033] In an embodiment of the present application, a first progress query request sent by the first terminal 110 is obtained, and the first progress query request includes a first task identifier taskId, which is used to query the first model training progress of the pre-trained language model corresponding to taskId. Optionally, the load balancing 121 can call the authentication routing 127 to complete the user authentication, obtain the training instance identifier tag_i after the user authentication is passed, and forward tag_i and the request content to the LLM training management module 122. The LLM training management module 122 forwards the task information such as taskId, tag_i and the training instance information to the seven-layer load 124, and the seven-layer load 124 calls the learning result query service 128 to query the first model training progress of the pre-trained language model. The learning result query service 128 queries the database 126 to obtain the first model training progress; based on the first model training progress, the model training progress information is generated; and the model training progress information is sent to the first terminal 110.

[0034] In an exemplary embodiment, if Figure 3 As shown, the LLM personalized training progress query method may include the following steps: Step 301: The first terminal 110 queries the model training progress corresponding to taskId; Step 302: The load balancer 121 completes the authentication and obtains the mapping relationship between the user ID and the training task ID tag. If the user ID has been bound to the training tag, the corresponding tag_i is returned; otherwise, a new training tag is assigned to it and the corresponding tag_i is returned; Step 303: The load balancer 121 queries the model training progress corresponding to taskId, with tag_i; Step 304: The LLM training management module 122 queries the database 126 for the model training progress corresponding to taskId through the seven-layer load 124 and the learning result query service 128; Step 305: the database 126 returns the model training progress corresponding to the taskId to the first terminal 110.

[0035] In a possible implementation, after the above step 203 allocates the model training data to the model learning service instance corresponding to the training instance identifier, the following further includes: By monitoring the training progress of the pre-trained language model, determining that the training progress is completed, sending the model training result to the first terminal, wherein the model training result is used to indicate that the text reasoning model has been trained.

[0036] In a specific implementation, the LLM learning service instance tag_i can call the learning progress processor 125 to complete the creation of the LLM training task, start the learning script to generate a learning log, and the learning progress processor 125 is used to parse the learning log corresponding to the training task and update the training progress to the database 126. By monitoring the training progress in the database 126, it is determined that the training progress is completed, and the model training result is sent to the first terminal 110. Among them, the above-mentioned monitoring of the training progress of the pre-trained language model includes: Obtain a learning log generated by the pre-trained language model during the training process; and obtain the training progress of the pre-trained language model by parsing the learning log.

[0037] In an embodiment of the present application, a model learning service instance, such as an LLM learning service tag_i, can call the learning progress processor 125 to complete the creation of an LLM training task, and then start the learning script to generate a learning log. The learning progress processor 125 is used to obtain the learning log generated by the LLM during the training process, parse the learning log corresponding to the training task, obtain the training progress, and update the training progress to the database 126.

[0038] In a possible implementation, the sending of the model training result to the first terminal includes: According to the configuration information of the first terminal, a prompt message corresponding to the configuration information is generated, wherein the prompt message includes the model training result; and the prompt message is sent to the first terminal.

[0039] In an embodiment of the present application, the model training result can be sent to the first terminal 110 by calling the message-must-reach platform 129. The message-must-reach platform 129 is used to generate a prompt message corresponding to the configuration information according to the configuration information of the first terminal 110, and send the prompt message to the first terminal 110. For example, if the recipient is a 5G message terminal, the message is sent based on platforms such as 5G messages; if the recipient is a non-5G message terminal, the message is converted into other message forms and sent to the recipient to ensure that it reaches the terminal used by the recipient. Here, the prompt message can also be sent to the third terminal 130. The third terminal 130 can be the same terminal as the first terminal, or it can be a different terminal from the first terminal. It can be set according to actual needs.

[0040] In an exemplary embodiment, if Figure 4a and Figure 4b As shown, taking the pre-trained language model as LLM as an example, the LLM training method may include the following steps: Step 401: The first terminal 110 uploads a training file and starts training; Step 402: The load balancer 121 completes the authentication and obtains the mapping relationship between the user ID and the training task ID tag. If the user ID has been bound to the training tag, the corresponding tag_i is returned; otherwise, a new training tag is assigned to it and the corresponding tag_i is returned; Step 403: The load balancer 121 starts the training task and forwards tag_i and the training file to the LLM training management module 122; Step 404: The LLM training management module 122 returns the first task identifier taskId to the first terminal 110, so that the first terminal 110 can query the model training progress according to the taskId; Step 405: The LLM training management module 122 calls the model training preprocessing tool 123 to complete the processing of the training file to the model training data; Step 406: passing the model training data and taskId, tag_i to the seventh-layer load 124; Step 407: The layer 7 load 124 parses tag_i and forwards it to the LLM learning service instance tag_i; Step 408: The LLM learning service instance tag_i generates a learning task; Step 409: the learning progress processor 125 analyzes the learning log corresponding to the learning task, creates a learning record and stores it in the database 126; Step 410: The LLM learning service instance tag_i starts the learning script to generate a learning log; Step 411: The learning progress processor parses the learning log, generates the learning progress and updates the training results; Step 412: When the learning is finished, the training result is pushed to the third terminal 130 by calling the message answering platform 129.

[0041] In this way, by preprocessing the training files through the model training preprocessing tool and assigning the preprocessed model training data to the corresponding model learning service instance for model training, the technical threshold for fine-tuning and training of pre-trained language models can be lowered, allowing ordinary users to train and fine-tune models more conveniently. By assigning model training data through training instance identification, multi-user isolated training can be achieved, ensuring that multiple users can perform model training at the same time, thereby improving the training efficiency of the text reasoning model. At the same time, by parsing the learning logs generated by the pre-trained language model during the training process and determining the training progress of the pre-trained language model, the transparency of the model training process can be enhanced. If problems such as training stagnation, insufficient resources, and incorrect configuration occur during the model training process, users can obtain feedback in real time and make timely adjustments or interventions to further improve the training efficiency of the text reasoning model.

[0042] Figure 5 The flowchart of the digital human text reasoning method provided in the embodiment of the present application is shown. The execution subject of the method can be a server, wherein the server can be a background server of a certain business, or a background server of a certain platform or application (for example, a smart question-and-answer platform, a digital human processing system, etc.). Figure 1 The AI ​​Agent 1210 in the digital human processing system is used as an example for explanation. As shown in the figure, the method 500 may include the following steps: Step 501: respond to a service request initiated by a second terminal, and obtain text content and a training instance identifier corresponding to the service request.

[0043] The second terminal 140 and the first terminal 110 may be the same terminal or two different terminals.

[0044] In a specific implementation, the user initiates a business request to the digital human processing system 120 through the second terminal 140, the message service 1211 obtains the business request, calls the authentication route 127 to complete the user authentication, and obtains the training instance identifier tag_i, and forwards the text content corresponding to the business request and the training instance identifier tag_i to the AI ​​Agent 1210.

[0045] Step 502: Inferring the text content through the text reasoning model corresponding to the training instance identifier to obtain a model reasoning result.

[0046] Wherein, the text reasoning model is trained according to the above-mentioned text reasoning model training method.

[0047] In a specific implementation, after obtaining the text content and the training instance identifier tag_i, the AI ​​Agent 1210 requests the text reasoning model corresponding to the training instance identifier through the seven-layer load 124. The text reasoning model reasons the text content by loading the corresponding model learning service instance to obtain the model reasoning result. For example, the text reasoning model is the LLM reasoning service tag_i, and the model learning service instance is the LLM learning service tag_i.

[0048] Optionally, AI Agent 1210 can also request the knowledge base 1212 to reason through the seven-layer load 124 and recall the query content. If the query content needs to further query the text reasoning model LLM, AI Agent 1210 further requests the corresponding LLM reasoning service tag_i through the seven-layer load 124.

[0049] Step 503: Generate a target video stream according to the model inference result and the preset digital human image.

[0050] The preset digital human image may be a digital human image pre-trained according to a video file provided by a user, or may be a default digital human image.

[0051] In a specific implementation, after obtaining the model inference result, the content-to-video stream inference request is initiated by calling the digital human terminal agent 1213. The digital human terminal agent 1213 calls the authentication route to complete the user authentication, obtains the training instance tag_i, and forwards tag_i and the request content to the media stream payload 1214. The media stream payload 1214 forwards the request to the digital human inference service tag_i. The digital human inference service tag_i completes the inference of the model inference result to the lip-sync video stream by presetting the digital human image, and generates the target video stream.

[0052] Step 504: Send the target video stream to the second terminal.

[0053] In a specific implementation, the target video stream is returned to the second terminal 140, and the second terminal 140 renders the corresponding target video stream into a digital human playing the LLM reasoning result.

[0054] In this way, through automated data preprocessing, training instance allocation, and training progress monitoring, the technical threshold for fine-tuning and training pre-trained language models can be lowered, allowing ordinary users to train and fine-tune models more conveniently. At the same time, by allocating model training data through training instance identification, multi-user isolated training can be achieved, ensuring that multiple users can train models at the same time, thereby improving the training efficiency of text reasoning models. In addition, ordinary users trigger corresponding text reasoning and video generation through business requests, avoiding cumbersome operations and in-depth understanding, which is suitable for widespread promotion to ordinary user groups.

[0055] In a possible implementation, in the above step 501, responding to the service request initiated by the second terminal and obtaining the text content and the training instance identifier corresponding to the service request includes: In response to a service request initiated by a second terminal, a second user identifier is obtained by performing user authentication on the second terminal; text content corresponding to the service request is obtained, and a training instance identifier corresponding to the text content is determined according to the second user identifier.

[0056] In an embodiment of the present application, after obtaining the service request initiated by the second terminal 140, the load balancing 121 in the CPU resource pool can call the authentication routing 127 to perform user authentication on the second terminal 110, obtain the second user identifier corresponding to the second terminal, obtain the text content corresponding to the service request, and determine the training instance identifier tag_i corresponding to the second user identifier based on the mapping relationship between the pre-stored user identifier and the training instance identifier, and determine tag_i as the training instance identifier corresponding to the text content.

[0057] Optionally, if the user identification ID is not bound to a corresponding training instance identifier, a new training instance identifier is allocated to the user identification ID.

[0058] In a possible implementation, in the above step 502, the text content is inferred by the text inference model corresponding to the training instance identifier to obtain a model inference result, including: The text content is inferred by the text reasoning model corresponding to the training instance identifier to obtain an initial model reasoning result; when it is determined that the initial model reasoning result includes a service call, the processing result obtained after the target service corresponding to the service call processes the text content is obtained, and the processing result and the text content are again inferred by the text reasoning model to obtain a model reasoning result.

[0059] In an embodiment of the present application, AI Agent 1210 requests the corresponding LLM reasoning service tag_i through the seven-layer load 124. The LLM reasoning service tag_i returns the initial model reasoning result. AI Agent 1210 determines whether there is a service call in the initial model reasoning result, for example, a functionCall call. If the initial model reasoning result includes functionCall, the corresponding functionCall is called to complete the processing of the corresponding task, and the processing result and text content are encapsulated in a prompt, and the LLM reasoning service tag_i is requested again for reasoning to obtain the model reasoning result.

[0060] In an exemplary embodiment, the above-mentioned digital human text reasoning method includes a text reasoning processing stage and a digital human personalized reasoning stage, such as Figure 6 As shown in , the text reasoning processing stage may include the following steps: Step 601: The second terminal 140 initiates a service request; Step 602: the message service 1211 obtains the service request, calls the authentication route 127 to complete the user authentication, and obtains the training instance identifier tag_i; Step 603: The message service 1211 forwards the text content corresponding to the business request and the training instance identifier tag_i to the AI ​​Agent 1210; Step 604: AI Agent 1210 requests reasoning from knowledge base 1212 via the seven-layer load 124; Step 605: the knowledge base 1212 searches for the corresponding database table according to tag_i, starts the conventional knowledge base recall process, recalls the query content, and returns the knowledge base recall result; Step 606: Continue LLM reasoning according to the knowledge base recall result; Step 6061: Request the LLM reasoning service tag_i corresponding to the training instance identifier through the seven-layer load 124 to obtain the initial model reasoning result; Step 6062: If there is functionCall in the initial model inference result, call the corresponding functionCall to complete the processing of the corresponding task, and encapsulate the processing result and text content in prompt; Step 6063: Request the LLM reasoning service tag_i to perform reasoning again and obtain the model reasoning result; Step 6064: Return the model inference result to the second terminal 140.

[0061] In an exemplary embodiment, if Figure 7 As shown in , the digital human personalized reasoning stage can include the following steps: Step 607: The second terminal 140 receives the model inference result; Step 608: The second terminal 140 initiates a content-to-video stream inference request to the digital human terminal agent 1213; Step 609: the digital human terminal agent 1213 calls the authentication route to complete the user authentication and obtain the training instance tag_i; Step 610: the digital human terminal agent 1213 forwards tag_i and the request content to the media stream payload 1214; Step 611: the media stream payload 1214 forwards the request to the digital human reasoning service tag_i; Step 612: the digital human reasoning service tag_i returns the target video stream; Step 613: The second terminal 140 renders the target video stream into a digital human playing the LLM reasoning result.

[0062] In a possible implementation, the preset digital human image is obtained in the following manner: In response to the digital human training request initiated by the second terminal, a video file corresponding to the digital human training request is obtained; according to a pre-configured digital human training preprocessing tool, the video file is preprocessed to obtain image training data; by assigning the image training data to the digital human learning service instance corresponding to the training instance identifier, a preset digital human image is obtained, wherein the digital human learning service instance is used to train a preset model according to the image training data to obtain the preset digital human image.

[0063] In an exemplary embodiment, if Figure 8a and Figure 8bAs shown, the digital human personalized image training method may include the following steps: Step 801: The second terminal 130 uploads a video file and starts training; Step 802: The load balancer 121 completes the authentication and obtains the mapping relationship between the user ID and the training task ID tag. If the user ID has been bound to the training tag, the corresponding tag_i is returned; otherwise, a new training tag is assigned to it and the corresponding tag_i is returned. Step 803: the load balancer 121 starts the training task and forwards the tag_i and the video file to the digital human image training management module 1215; Step 804: the digital human image training management module 1215 returns the first task identifier taskId to the first terminal 110, so that the first terminal 110 can query the model training progress according to the taskId; Step 805: the digital human image training management module 1215 calls the digital human training pre-processing tool 1216 to complete the processing of the video file into image training data; Step 806: passing the image training data and taskId, tag_i to the seven-layer load 124; Step 807: the seventh layer load 124 parses tag_i and forwards it to the digital human learning service instance tag_i; Step 808: The digital human learning service instance tag_i generates a learning task, and the preset model is trained according to the image training data through the digital human learning service instance tag_i to obtain a preset digital human image.

[0064] Wherein, after obtaining the preset digital human image by assigning the image training data to the digital human learning service instance corresponding to the training instance identifier, the method further includes: By monitoring the training progress of the preset model, it is determined that the training progress is completed, and the digital human training result is sent to the second terminal, wherein the digital human training result is used to indicate that the preset digital human image has been trained.

[0065] Continuing with the above embodiment, the above method further includes: Step 809: the learning progress processor 125 analyzes the learning log corresponding to the learning task, creates a learning record and stores it in the database 126; Step 810: the digital human learning service instance tag_i starts the learning script to generate a learning log; Step 811: The learning progress processor analyzes the learning log, generates the learning progress and updates the digital human training results; Step 812: When the learning is finished, the digital human training results are pushed to the third terminal 130 by calling the message answering platform 129.

[0066] In a possible implementation, the above-mentioned allocating the image training data to the digital human learning service instance corresponding to the training instance identifier includes: Creating a digital human training task, and determining a second task identifier corresponding to the digital human training task; The second task identifier and the image training data are routed to the digital human learning service instance corresponding to the training instance identifier.

[0067] In the embodiment of the present application, after acquiring the image training data and tag_i, the digital human image training management module 1215 creates a model training task, determines the second task identifier digital human taskId corresponding to the model training task, and sends the image training data, digital human taskId and tag_i to the seven-layer load 124. The seven-layer load 124 parses tag_i and routes it to the corresponding digital human learning service instance tag_i.

[0068] In a possible implementation, after routing the second task identifier and the image training data to the digital human learning service instance corresponding to the training instance identifier, the method further includes: Obtain a second progress query request sent by the second terminal, wherein the second progress query request includes a second task identifier; obtain a second model training progress of the preset model according to the second task identifier; generate digital human image training progress information corresponding to the second progress query request according to the second model training progress; and send the digital human image training progress information to the second terminal.

[0069] In an exemplary embodiment, if Fig. 9 As shown, the above-mentioned digital personalized training progress query method may include the following steps: Step 901: the second terminal 140 queries the model training progress corresponding to taskId; Step 902: The load balancer 121 completes the authentication and obtains the mapping relationship between the user ID and the training task ID tag. If the user ID has been bound to the training tag, the corresponding tag_i is returned; otherwise, a new training tag is assigned to it and the corresponding tag_i is returned. Step 903: The load balancer 121 queries the model training progress corresponding to taskId, carrying tag_i; Step 904: the digital human image training management module 1215 queries the database 126 for the digital human image training progress corresponding to taskId through the seven-layer load 124 and the learning result query service 128; Step 905 : the database 126 returns the digital human image training progress corresponding to the taskId to the second terminal 140 .

[0070] In this way, by analyzing the learning logs generated by the preset model during the training process and determining the training progress of the preset model, the transparency of the model training process can be enhanced. If problems such as training stagnation, insufficient resources, and incorrect configuration occur during the model training process, users can obtain real-time feedback and make timely adjustments or interventions to further improve the training efficiency of the digital human image.

[0071] Fig.10 The structure diagram of the text reasoning model training device provided in the embodiment of the present application is shown. The text reasoning model training device can be implemented as follows: Figure 2 All or part of the contents of the illustrated embodiment, the text reasoning model training device 1000 includes: The file acquisition module 1010 is used to respond to the model training request initiated by the first terminal and acquire the training file and training instance identifier corresponding to the model training request; A file processing module 1020 is used to pre-process the training file according to a pre-configured model training pre-processing tool to obtain model training data; The model training module 1030 is used to obtain a text reasoning model obtained by training a pre-trained language model according to the model training data by the model learning service instance by assigning the model training data to the model learning service instance corresponding to the training instance identifier.

[0072] In a possible implementation, the file acquisition module 1010, when used to respond to the model training request initiated by the first terminal and obtain the training file and the training instance identifier corresponding to the model training request, is specifically used to: In response to a training request initiated by the first terminal, obtaining a first user identifier corresponding to the first terminal by performing user authentication on the first terminal; A training file corresponding to the training request is obtained, and a training instance identifier corresponding to the training file is determined according to the first user identifier.

[0073] In a possible implementation, the model training module 1030, when used to allocate the model training data to the model learning service instance corresponding to the training instance identifier, is specifically used to: Creating a model training task, and determining a first task identifier corresponding to the model training task; The first task identifier and the model training data are routed to the model learning service instance corresponding to the training instance identifier.

[0074] In a possible implementation, the model training module 1030 is further used to: Acquire a first progress query request sent by the first terminal, where the first progress query request includes a first task identifier; According to the first task identifier, obtaining a first model training progress of the pre-trained language model; Generate model training progress information corresponding to the training progress query request according to the first model training progress; Send the model training progress information to the first terminal.

[0075] In a possible implementation, the text reasoning model training device 1000 further includes: The progress monitoring module 1040 is used to monitor the training progress of the pre-trained language model, determine that the training progress is completed, and send the model training result to the first terminal, wherein the model training result is used to indicate that the text reasoning model has been trained. In a possible implementation, the progress monitoring module 1040, when used to monitor the training progress of the pre-trained language model, is specifically used to: Obtaining a learning log generated by the pre-trained language model during the training process; The training progress of the pre-trained language model is obtained by parsing the learning log.

[0076] In a possible implementation, the progress monitoring module 1040, when used to send the model training result to the first terminal, is specifically used to: Generate, according to the configuration information of the first terminal, a prompt message corresponding to the configuration information, wherein the prompt message includes the model training result; Send the prompt message to the first terminal.

[0077] The embodiment of the present application provides a text reasoning model training device, including a file acquisition module, a file processing module and a model training module; the file acquisition module responds to the model training request initiated by the first terminal, and obtains the training file corresponding to the model training request; the file processing module preprocesses the training file according to the pre-configured model training preprocessing tool to obtain the model training data; the model training module assigns the model training data to the model learning service instance corresponding to the training instance identifier, and obtains the text reasoning model obtained by the model learning service instance training the pre-trained language model according to the model training data. In this way, the training file is preprocessed by the model training preprocessing tool, and the preprocessed model training data is assigned to the corresponding model learning service instance for model training, which can simplify the fine-tuning and training process of the pre-trained language model, reduce the complexity of operation, and make the model application easier to promote. At the same time, by allocating model training data through the training instance identifier, multi-user isolation training can be achieved, ensuring that multiple users perform model training at the same time, thereby improving the training efficiency of the text reasoning model.

[0078] Fig.11 The structure diagram of the digital human text reasoning device provided in the embodiment of the present application is shown. The digital human text reasoning device can be implemented as follows: Figure 5 All or part of the contents of the embodiment shown, the digital human text reasoning device 1100, includes: The service request module 1110 is used to respond to the service request initiated by the second terminal and obtain the text content and training instance identifier corresponding to the service request; A model reasoning module 1120 is used to reason about the text content by identifying the text reasoning model corresponding to the training instance to obtain a model reasoning result; wherein the text reasoning model is trained according to the method as described above; The video stream generation module 1130 is used to generate a target video stream according to the model inference result and the preset digital human image; The video stream sending module 1140 is used to send the target video stream to the second terminal.

[0079] In a possible implementation, the service request module 1110, when used to respond to the service request initiated by the second terminal and obtain the text content and the training instance identifier corresponding to the service request, is specifically used to: In response to a service request initiated by the second terminal, obtaining a second user identifier by performing user authentication on the second terminal; The text content corresponding to the business request is obtained, and the training instance identifier corresponding to the text content is determined according to the second user identifier.

[0080] In a possible implementation, the model reasoning module 1120, when used to identify the text reasoning model corresponding to the training instance, reason on the text content, and obtain the model reasoning result, is specifically used to: Inferring the text content through the text reasoning model corresponding to the training instance identifier to obtain an initial model reasoning result; When it is determined that the initial model inference result includes a service call, the processing result obtained after the target service corresponding to the service call processes the text content is obtained, and the processing result and the text content are again inferred through the text inference model to obtain a model inference result.

[0081] In a possible implementation, the video stream generation module 1130 is configured to obtain the preset digital human image in the following manner: Responding to the digital human training request initiated by the second terminal, obtaining a video file corresponding to the digital human training request; Preprocessing the video file according to a preconfigured digital human training preprocessing tool to obtain image training data; The preset digital human image is obtained by assigning the image training data to the digital human learning service instance corresponding to the training instance identifier, wherein the digital human learning service instance is used to train a preset model according to the image training data to obtain the preset digital human image.

[0082] In a possible implementation, the video stream generating module 1130 is further configured to: By monitoring the training progress of the preset model, it is determined that the training progress is completed, and the digital human training result is sent to the second terminal, wherein the digital human training result is used to indicate that the preset digital human image has been trained.

[0083] In a possible implementation, the video stream generation module 1130, when used to allocate the image training data to the digital human learning service instance corresponding to the training instance identifier, is specifically used to: Creating a digital human training task, and determining a second task identifier corresponding to the digital human training task; The second task identifier and the image training data are routed to the digital human learning service instance corresponding to the training instance identifier.

[0084] In a possible implementation, the video stream generating module 1130 is further configured to: Acquire a second progress query request sent by the second terminal, where the second progress query request includes a second task identifier; According to the second task identifier, obtaining a second model training progress of the preset model; Generate digital human image training progress information corresponding to the second progress query request according to the second model training progress; The digital human image training progress information is sent to the second terminal.

[0085] The embodiment of the present application provides a digital human text reasoning device, including a business request module, a model reasoning module, a video stream generation module and a video stream sending module; the business request module responds to the business request initiated by the second terminal, obtains the text content and the training instance identifier corresponding to the business request; the model reasoning module uses the text reasoning model corresponding to the training instance identifier to reason the text content and obtain the model reasoning result; the video stream generation module generates a target video stream according to the model reasoning result and the preset digital human image; the video stream sending module sends the target video stream to the second terminal. In this way, the training file is preprocessed by the model training preprocessing tool, and the preprocessed model training data is assigned to the corresponding model learning service instance for model training, which can simplify the fine-tuning and training process of the pre-trained language model, reduce the complexity of operation, and make the model application easier to promote. In addition, ordinary users trigger corresponding text reasoning and video generation through business requests, avoiding cumbersome operations and in-depth understanding, and are suitable for wide promotion to ordinary user groups.

[0086] Fig.12 The hardware structure diagram of the electronic device provided by the embodiment of the present application is shown. Referring to the figure, at the hardware level, the electronic device 1200 includes a processor 1210, and optionally, includes an internal bus 1220, a network interface 1230, and a memory 1240. Among them, the memory 1240 may include a memory 1241, such as a high-speed random access memory (Random-Access Memory, RAM), and may also include a non-volatile memory 1242 (non-volatile memory), such as at least one disk storage, etc. Of course, the electronic device 900 may also include hardware required for other services.

[0087] The processor 1210, the network interface 1230 and the memory may be interconnected via an internal bus 1220, which may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0088] The memory 1240 stores programs. Specifically, the programs may include program codes, and the program codes include computer operation instructions. The memory 1240 may include a memory 1241 and a non-volatile memory 1242 , and provide instructions and data to the processor 1210 .

[0089] The processor 1210 reads the corresponding computer program from the non-volatile memory 1242 into the memory and then runs it, forming a device for locating the target user at the logical level. The processor 1210 executes the program stored in the memory and specifically performs: Figure 2 or Figure 5 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0090] The above application Figure 2 or Figure 5The method disclosed in the illustrated embodiment can be applied to the processor 1210 or implemented by the processor 1210. The processor 1210 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 1210. The above processor 1210 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0091] The computer device can also execute the methods described in the foregoing method embodiments, and realize the functions and beneficial effects of the methods described in the foregoing method embodiments, which will not be repeated here.

[0092] Of course, in addition to software implementation, the electronic device 1200 of the present application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0093] The embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable medium stores one or more programs, and when the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes Figure 2 or Figure 5 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0094] The computer-readable storage medium includes a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0095] Furthermore, an embodiment of the present application also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the following process is implemented: Figure 2 or Figure 5 The method disclosed in the illustrated embodiment realizes the functions and beneficial effects of each method described in the foregoing method embodiments, which will not be described in detail here.

[0096] The embodiments of the present application can be applied to various electronic device collaboration or interconnection scenarios, including: collaboration and interconnection between mobile phones and laptops / tablet computers; collaboration and interconnection between mobile terminals and smart TVs / displays; collaboration and interconnection between mobile phones or tablet computers and in-car entertainment systems; collaboration and interconnection between mobile terminals and smart conference systems, etc., so as to meet the needs of users in various scenarios such as smart homes, smart offices, and smart travel.

[0097] In short, the above is only a preferred embodiment of the present application and does not limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0098] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0099] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0100] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0101] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A text reasoning model training method, characterized in that: include: Responding to the model training request initiated by the first terminal, obtaining a training file and a training instance identifier corresponding to the model training request; Preprocessing the training file according to a preconfigured model training preprocessing tool to obtain model training data; By allocating the model training data to the model learning service instance corresponding to the training instance identifier, a text inference model is obtained by the model learning service instance training the pre-trained language model according to the model training data.

2. The method according to claim 1, characterized in that The step of responding to the model training request initiated by the first terminal and obtaining a training file and a training instance identifier corresponding to the model training request includes: In response to a training request initiated by the first terminal, obtaining a first user identifier corresponding to the first terminal by performing user authentication on the first terminal; A training file corresponding to the training request is obtained, and a training instance identifier corresponding to the training file is determined according to the first user identifier.

3. The method according to claim 1, characterized in that The allocating the model training data to the model learning service instance corresponding to the training instance identifier includes: Creating a model training task, and determining a first task identifier corresponding to the model training task; The first task identifier and the model training data are routed to the model learning service instance corresponding to the training instance identifier.

4. The method according to claim 3, characterized in that After routing the first task identifier and the model training data to the model learning service instance corresponding to the training instance identifier, the method further includes: Acquire a first progress query request sent by the first terminal, where the first progress query request includes a first task identifier; According to the first task identifier, obtaining a first model training progress of the pre-trained language model; Generate model training progress information corresponding to the training progress query request according to the first model training progress; Send the model training progress information to the first terminal.

5. The method according to claim 1, characterized in that After allocating the model training data to the model learning service instance corresponding to the training instance identifier, the method further includes: By monitoring the training progress of the pre-trained language model, determining that the training progress is completed, sending the model training result to the first terminal, wherein the model training result is used to indicate that the text reasoning model has been trained.

6. The method according to claim 5, characterized in that The monitoring the training progress of the pre-trained language model includes: Obtaining a learning log generated by the pre-trained language model during the training process; The training progress of the pre-trained language model is obtained by parsing the learning log.

7. A digital human text reasoning method, characterized in that: include: Responding to a service request initiated by the second terminal, obtaining text content and a training instance identifier corresponding to the service request; The text content is inferred by the text reasoning model corresponding to the training instance identifier to obtain a model reasoning result; wherein the text reasoning model is trained according to the method according to any one of claims 1 to 6; Generate a target video stream according to the model inference result and the preset digital human image; The target video stream is sent to the second terminal.

8. The method according to claim 7, characterized in that The responding to the service request initiated by the second terminal and obtaining the text content and the training instance identifier corresponding to the service request includes: In response to a service request initiated by the second terminal, obtaining a second user identifier by performing user authentication on the second terminal; The text content corresponding to the business request is obtained, and the training instance identifier corresponding to the text content is determined according to the second user identifier.

9. The method according to claim 7, characterized in that: The text reasoning model corresponding to the training instance identifier is used to reason about the text content to obtain a model reasoning result, including: Inferring the text content through the text reasoning model corresponding to the training instance identifier to obtain an initial model reasoning result; When it is determined that the initial model inference result includes a service call, the processing result obtained after the target service corresponding to the service call processes the text content is obtained, and the processing result and the text content are again inferred through the text inference model to obtain a model inference result.

10. The method according to claim 7, characterized in that The preset digital human image is obtained by: Responding to the digital human training request initiated by the second terminal, obtaining a video file corresponding to the digital human training request; Preprocessing the video file according to a preconfigured digital human training preprocessing tool to obtain image training data; The preset digital human image is obtained by assigning the image training data to the digital human learning service instance corresponding to the training instance identifier, wherein the digital human learning service instance is used to train a preset model according to the image training data to obtain the preset digital human image.

11. The method according to claim 10, characterized in that After obtaining the preset digital human image by assigning the image training data to the digital human learning service instance corresponding to the training instance identifier, the method further includes: By monitoring the training progress of the preset model, it is determined that the training progress is completed, and the digital human training result is sent to the second terminal, wherein the digital human training result is used to indicate that the preset digital human image has been trained.

12. The method according to claim 10, characterized in that The step of allocating the image training data to the digital human learning service instance corresponding to the training instance identifier includes: Creating a digital human training task, and determining a second task identifier corresponding to the digital human training task; The second task identifier and the image training data are routed to the digital human learning service instance corresponding to the training instance identifier.

13. The method according to claim 12, characterized in that After routing the second task identifier and the image training data to the digital human learning service instance corresponding to the training instance identifier, the method further includes: Acquire a second progress query request sent by the second terminal, where the second progress query request includes a second task identifier; According to the second task identifier, obtaining a second model training progress of the preset model; Generating digital human image training progress information corresponding to the second progress query request according to the second model training progress; The digital human image training progress information is sent to the second terminal.

14. A text reasoning model training device, characterized in that: include: A file acquisition module, used to respond to the model training request initiated by the first terminal, and acquire the training file and training instance identifier corresponding to the model training request; A file processing module, used to pre-process the training file according to a pre-configured model training pre-processing tool to obtain model training data; The model training module is used to obtain a text reasoning model obtained by training a pre-trained language model according to the model training data by the model learning service instance by assigning the model training data to the model learning service instance corresponding to the training instance identifier.

15. A digital human text reasoning device, characterized in that: include: A service request module, used to respond to a service request initiated by the second terminal, and obtain text content and a training instance identifier corresponding to the service request; A model reasoning module, used to reason the text content through the text reasoning model corresponding to the training instance identifier to obtain a model reasoning result; wherein the text reasoning model is trained according to the method according to any one of claims 1 to 6; A video stream generation module, used to generate a target video stream according to the model inference result and a preset digital human image; A video stream sending module is used to send the target video stream to the second terminal.

16. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method according to any one of claims 1 to 13 are implemented.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 13 are implemented.

18. A computer program product, characterized in that The computer program product comprises a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer is caused to perform the steps of the method according to any one of claims 1 to 13.

Images