Data transmission method, dialog response method, and data transmission apparatus

Abstract

The application discloses a data transmission method, a dialogue response method and a data transmission device. The data transmission method comprises the following steps: obtaining transmission data of a sender; obtaining model parameters corresponding to the transmission data based on the transmission data and a first model; and transmitting the model parameters to a receiver, wherein the model parameters are used to load a second model of the receiver so that the second model can answer a question for the transmission data; and the first model and the second model are the same model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data transmission method, a dialogue response method, and a data transmission device. Background Technology

[0002] In traditional data transmission, the sender transmits the data to the receiver through a data transmission channel. This method is too simplistic. Summary of the Invention

[0003] This application provides a data transmission method, including: obtaining transmission data from a sender; obtaining model parameters corresponding to the transmission data based on the transmission data and a first model; transmitting the model parameters to a receiver, wherein the model parameters are used to load a second model of the receiver so that the second model can answer questions about the transmission data; wherein the first model and the second model are the same model.

[0004] In some embodiments, obtaining model parameters corresponding to the transmission data based on the transmission data and the first model includes: obtaining a first matrix, the first matrix being a key matrix; obtaining a second matrix based on the first model using the transmission data and the first matrix, the second matrix representing the encrypted transmission data; wherein the second matrix is ​​used to transmit to a receiver so that the receiver obtains model parameters corresponding to the transmission data by multiplying the second matrix by a third matrix, and the first matrix is ​​the same as the third matrix.

[0005] In some embodiments, the first matrix is ​​one of two matrices corresponding to the incremental model parameters obtained by training the first model based on the target file set with the weight parameters frozen; wherein, during each round of data transmission, the parameters of the first matrix remain unchanged; and the transmitted data does not belong to the target file set.

[0006] In some embodiments, different sets of target files correspond to different first matrices.

[0007] In some embodiments, the method further includes updating the first matrix based on the change in the number of files in the target file set.

[0008] This application embodiment also provides a dialogue response method, including: obtaining an identifier of a receiver; configuring a response method for responding to a dialogue request from the receiver's identifier based on the receiver's identifier; the response method is to send model parameters to the receiver's identifier, the model parameters being loaded into a second model that the receiver can enable so that the second model can answer questions for the sender; the model parameters are model parameters corresponding to the target data set generated by a first model that the sender can enable based on the sender's target data set.

[0009] This application embodiment also provides a dialogue response method, including: obtaining question information sent by an identifier of a receiver; determining a response method for the question information; in response to the response method, sending model parameters to the identifier of the receiver, wherein the model parameters are used to load a second model that the receiver can enable so that the second model can answer questions for the sender; the model parameters are model parameters corresponding to the target data set generated by the sender based on the target data set of the sender, which is enabled by a first model.

[0010] In some embodiments, determining the response method for responding to the question information includes: switching from a first response method to a second response method, wherein the first response method is to obtain user input information for the question information as a response to the question information; and the second response method is to send model parameters to the identifier of the receiver, wherein the model parameters are used to load a second model that the receiver can enable so that the second model can answer the question information for the sender.

[0011] This application embodiment also provides a dialogue response method, including: obtaining question information sent by a receiver's identifier; displaying the question information in a chat window based on the receiver's identifier; and, if the question information meets target conditions, having a smart assistant reply to the question information; wherein, the smart assistant's reply to the question information is generated by loading model parameters through a first model to reply to the question information; the model parameters are model parameters that the sender can enable the first model to generate based on the sender's target data set, corresponding to the target data set.

[0012] This application embodiment also provides a data transmission device, including: a first obtaining module for obtaining transmission data from a sender; a second obtaining module for obtaining model parameters corresponding to the transmission data based on the transmission data and a first model; and a transmission module for transmitting the model parameters to a receiver, wherein the model parameters are loaded into a second model of the receiver so that the second model can answer questions about the transmission data; wherein the first model and the second model are the same model. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a data transmission method according to an embodiment of this application;

[0015] Figure 2 This is a flowchart illustrating the process of obtaining the second matrix according to an embodiment of this application; Figure 3 The flow of the dialogue response method in this application embodiment Figure 1 ; Figure 4 The flow of the dialogue response method in this application embodiment Figure 2 ; Figure 5 The flow of the dialogue response method in this application embodiment Figure 3 ; Figure 6 This is a structural block diagram of a data transmission device according to an embodiment of this application. Detailed Implementation

[0016] Various embodiments and features of this application are described herein with reference to the accompanying drawings.

[0017] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.

[0018] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0019] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0020] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application.

[0021] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0022] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.

[0023] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0024] like Figure 1 As shown, the data transmission method includes the following steps: Step S101: Obtain the transmission data from the sender.

[0025] In this embodiment, the sender and receiver can be different electronic devices, each corresponding to a different user. The transmitted data can include one or more forms such as text, images, video, and audio. The entity executing the data transmission method can be the sender or a server between the sender and receiver.

[0026] The system can trigger the retrieval of the sender's transmitted data in response to a query regarding the transmitted data. For example, if the transmitted data consists of multiple year-end reports, and if the year-end reports include a target year-end report, the query could be "What does the summary of the target year-end report include?". It can also trigger the retrieval of the sender's transmitted data in response to a data transmission request from the receiver. Furthermore, it can trigger the retrieval of the sender's transmitted data in response to the fulfillment of target data transmission conditions (such as the sender receiving new data or reaching a specified data transmission time).

[0027] Step S102: Obtain model parameters corresponding to the transmission data based on the transmission data and the first model.

[0028] In this embodiment, the first model can be a neural network model or a large language model. The parameters of the first model can be adjusted based on the transmitted data (for example, the transmitted data can be used as sample data to train the first model). The adjusted parameters in the first model (such as the adjusted weights, biases, etc.) are determined as the model parameters corresponding to the transmitted data.

[0029] Step S103: Transmit the model parameters to the receiver. The model parameters are used to load a second model into the receiver so that the second model can answer questions about the transmitted data.

[0030] In this embodiment, the first model and the second model are the same model; for example, the first model and the second model can be the same base large language model. The model parameters are transmitted to the receiver, and the receiver can then load the model parameters into the second model, enabling the second model to possess knowledge corresponding to the transmitted data. When the receiver receives a question about the transmitted data, it can use the second model to deduce the corresponding answer, thereby allowing the receiver to reconstruct part or all of the transmitted data.

[0031] The method in this application involves the sender training a model based on the data to be transmitted, enabling the model to learn the data, which then becomes the model's learned knowledge. When the model's parameters are transmitted to the receiver, the receiver's model, based on these parameters, can answer any questions the receiver may have regarding the data to be transmitted. In other words, this application's embodiments change the way data is transmitted, increasing the diversity of data transmission. For example, it's similar to the sender teaching a postman the data to be transmitted; upon arrival at the receiver, the postman answers any questions the receiver may have regarding the data based on the learned knowledge. In contrast, existing technologies simply provide the data to the postman, who then leaves it with the receiver. The receiver then receives the data to be transmitted.

[0032] For example, if the transmitted data is a collection of documents including a first text (2025 Annual Work Report) and a first video, then if the question regarding the collection of documents is "What is the content of the first text in the collection of documents?", the receiver can use the second model to output the content of the first text. If the question regarding the collection of documents is "What content is included in the collection of documents?", the receiver can use the second model to output the first text and the first video. If the receiver's question regarding the collection of documents is "Summarize the key points of the 2025 Annual Work Report", then the receiver can reply based on the second model and the obtained model parameters with "Summarize the key points of the 2025 Annual Work Report including the following four items: XX; XX; XX."

[0033] The data transmission method of this application embodiment obtains transmission data from the sender; obtains model parameters corresponding to the transmission data based on the transmission data and a first model; transmits the model parameters to the receiver, and the model parameters are used to load a second model in the receiver so that the second model can answer questions about the transmission data; wherein the first model and the second model are the same model. In this way, the transmission data is converted into model parameters through the first model, and the model parameters are transmitted to the receiver. The receiver can load the model parameters into the second model, enabling the second model to reconstruct the transmission data. Since the transmission data is not directly transmitted and complex key management is not required, data transmission security is improved while computational overhead is reduced.

[0034] In some embodiments of this application, the step of obtaining model parameters corresponding to the transmission data based on the transmission data and the first model is as follows: Figure 2 As shown, it includes the following steps: Step S1021: Obtain the first matrix, which is a key matrix.

[0035] In this embodiment, the first matrix can be a pre-set key matrix, and the receiver stores a third matrix identical to the first matrix. The first matrix can be set locally on the sender's end, and the third matrix can be set locally on the receiver's end.

[0036] Step S1022: Based on the first model, a second matrix is ​​obtained using the transmitted data and the first matrix, wherein the second matrix represents the encrypted transmitted data.

[0037] In this embodiment, the parameters of the first model can be adjusted using the transmitted data to determine the model parameters corresponding to the transmitted data. The second matrix is ​​determined using the model parameters corresponding to the transmitted data and the first matrix. The second matrix represents the encrypted transmitted data. Subsequently, the second matrix can be transmitted to the receiver, so that the receiver can obtain the model parameters corresponding to the transmitted data by multiplying the third matrix and the second matrix, thereby realizing the transmission of model parameters to the receiver.

[0038] By determining the second matrix, only the second matrix can be transmitted to the receiver. Compared to directly transmitting model parameters, this reduces the amount of data transmitted and improves data transmission efficiency. Furthermore, since the third and second matrices need to be used simultaneously to determine the model parameters corresponding to the transmitted data, encrypted data transmission is achieved, improving the security of the data transmission process.

[0039] In some embodiments of this application, the first matrix is ​​one of two matrices corresponding to the incremental model parameters obtained by training the first model based on the target file set with the weight parameters frozen; In each round of data transmission, the parameters of the first matrix remain unchanged; the transmitted data does not belong to the target document set.

[0040] In this embodiment, the first model includes an incremental model. The weight parameters of the first model can be frozen, and the incremental model can be trained based on the target file set to obtain two matrices corresponding to the incremental model parameters. The product of these two matrices is the incremental model parameter. One of these two matrices is determined as the first matrix.

[0041] Optionally, incremental model parameters can be obtained by training on all data in the target file set, or by training on a portion of the data in the target file set or data of a specific type.

[0042] A single round of data transmission can be achieved by executing steps S101-S103 once. During each round of data transmission, the parameters of the first matrix remain unchanged; therefore, the second matrix changes as the transmitted data changes, enabling encrypted transmission of different data. The transmitted data may not belong to the target file set. For example, if the sender obtains new data outside the target file set, it may identify this new data as the transmitted data. Thus, even if an attacker obtains the target file set, they cannot determine the transmitted data, further enhancing the security of the data transmission process.

[0043] In some embodiments of this application, the transmitted data may also belong to the target file set. For example, the sender may identify historical files in the target file set as the transmitted data. Since the model parameters loaded into the second model can only be determined through the third matrix and the second matrix, even if the attacker obtains the second matrix, he cannot determine the model parameters without knowing the third matrix (or the first matrix), thereby improving the security and flexibility of the data transmission process.

[0044] In some embodiments of this application, the incremental model parameters can be the parameter increments ΔW of LoRA (low-rank adaptive model), where ΔW = A·B, the matrix size of ΔW is d×d, d is the hidden layer dimension of the first model, the matrix size of A is r×d, and the matrix size of B is d×r, where r is much smaller than d. Therefore, if the first matrix is ​​A, the second matrix B can be determined using the least squares method. B = argmin ||ΔW - A·B||_F² (size d×r); Here, argmin represents minimization, and ||.||_F² represents the square of the Frobenius norm, i.e., finding a matrix B such that the difference between A·B and ΔW is minimized.

[0045] In this way, the first and second matrices are obtained through LoRA. The rank of the second matrix is ​​much smaller than the rank of the incremental model parameters, which greatly reduces the amount of data transmitted (the amount of data transmitted is only 1 / r of the complete LoRA model. For example, if the first model uses Llama-2-7B, the amount of data transmitted can be reduced by 87.5% when r=8), which further reduces the transmission overhead and improves the data transmission efficiency.

[0046] Alternatively, as an alternative, a first matrix can be randomly generated based on the size of the target matrix. For example, if the target matrix has a size of r×d, a first matrix of size r×d can be randomly generated.

[0047] Alternatively, the size of the first matrix can be d×r, and the size of the second matrix can be r×d.

[0048] In some embodiments of this application, the first matrix can be stored in the sender's encrypted storage area (such as an HSM hardware module), and the third matrix can be stored in the receiver's encrypted storage area (such as an HSM hardware module), thereby further improving the security of the first matrix.

[0049] In some embodiments of this application, different sets of target files correspond to different first matrices.

[0050] In this embodiment, different target file sets can correspond to different file types, and can also correspond to different users or different user permission levels. For example, target file set 1 may be a collection of text files, target file set 2 a collection of image files, and target file set 3 a collection of video files. In this case, target file set 1, target file set 2, and target file set 3 each correspond to a different first matrix, thereby associating each first matrix with its corresponding file type. Alternatively, if target file set 1 corresponds to a first user, target file set 2 to a second user, and target file set 3 to a third user, then target file set 1, target file set 2, and target file set 3 each correspond to a different first matrix, thereby associating each first matrix with its corresponding user, thus improving the security of data transmission in multi-user scenarios.

[0051] By assigning different first matrices to different sets of target files, each set of target files can have its own unique first matrix, further improving the security of the data transmission process.

[0052] In some embodiments of this application, the data transmission method further includes the following steps: The first matrix is ​​updated based on the change in the number of files in the target file set.

[0053] In this embodiment, the target file set can change the number of files as new files are added, files are deleted, and cached files accumulate. The first matrix can be updated when the change in the number of files reaches the target data volume or when the change in the number of files in the target file set reaches the target proportion, thereby further improving the security of the data transmission process.

[0054] It is understandable that after updating the first matrix, the updated first matrix is ​​sent to the receiver, so that the receiver can replace the third matrix based on the updated first matrix, thereby updating the third matrix.

[0055] In some embodiments of this application, the data transmission method further includes the following steps: determining the existence duration of the first matrix, and updating the first matrix when the existence duration reaches the target duration, thereby further improving the security of the data transmission process.

[0056] In some embodiments of this application, after obtaining the second matrix based on the first model using the transmitted data and the first matrix, the process may further include: encrypting the second matrix using a shared key with the receiver to obtain an encrypted second matrix; and transmitting the encrypted second matrix to the receiver. After obtaining the encrypted second matrix, the receiver can decrypt it using the shared key to obtain the second matrix, thereby further improving the security of the second matrix transmission process.

[0057] This application also proposes a dialogue response method, such as... Figure 3 As shown, it includes the following steps: Step S201: Obtain the identifier of the recipient.

[0058] In this embodiment, the entity executing the dialogue response method can be the receiver or a server between the sender and receiver. The receiver and sender can engage in dialogue; the sender and receiver can be different electronic devices, each corresponding to a different user. The receiver's identifier can be the identifier of the corresponding electronic device or the identifier of the corresponding user; different receivers correspond to different identifiers. For example, the receiver's identifier can be any of the following forms: device identification code of the electronic device, user ID, user nickname, etc. The receiver's identifier is obtained before initiating the dialogue.

[0059] Step S202: Based on the identifier of the receiver, configure the response method for responding to the identifier dialogue request of the receiver.

[0060] In this embodiment, during the dialogue between the sender and receiver, the receiver's identifier can be used to send a dialogue request. After obtaining the receiver's identifier, the response method for responding to the dialogue request is configured based on the receiver's identifier, so that the receiver responds to the dialogue request in the subsequent dialogue process using this response method.

[0061] The sender can enable a first model to generate model parameters corresponding to the sender's target data set. For example, the sender can adjust the parameters of the first model using the target data set, and determine the adjusted parameters in the first model as the model parameters corresponding to the target data set. The response method is to send the model parameters to the receiver as an identifier. The model parameters are used to load a second model that the receiver can enable, enabling the second model to answer questions from the sender. The first and second models are the same model; for example, the first and second models can be the same base large language model.

[0062] In this response mode, if a dialogue request is received from the recipient's identifier, model parameters are sent to the recipient's identifier. The recipient loads the model parameters into the second model, enabling the second model to possess knowledge corresponding to the target dataset. After receiving a question from the sender regarding the target dataset, the second model can accurately deduce the corresponding answer.

[0063] The dialogue response method of this application embodiment obtains the question information sent by the receiver's identifier; determines the response method for the question information; and, in response to the response method, sends model parameters to the receiver's identifier. The model parameters are used to load a second model that the receiver can enable, enabling the second model to answer questions from the sender. The model parameters are generated by the sender based on the sender's target data set, using a first model that is also enabled. By configuring the response method in conjunction with the receiver's identifier, the receiver can accurately answer the sender's questions using the second model, improving the dialogue efficiency between the sender and receivers corresponding to different identifiers. Since the second model possesses knowledge of the target data set by loading the model parameters, it can transmit the target data set based on the model parameters corresponding to the target data set without complex key management, thereby improving data transmission security while reducing computational overhead.

[0064] In some embodiments of this application, the process by which the sender enables a first model to generate model parameters corresponding to the target data set based on the sender's target data set includes: obtaining a first matrix, which is a key matrix; obtaining a second matrix based on the first model, using the target data set and the first matrix, whereby the second matrix represents the encrypted target data set; wherein the second matrix is ​​used to transmit to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by a third matrix, and the first matrix and the third matrix are identical. The identifier for sending the model parameters to the receiver includes: sending the identifier of the second matrix to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by the third matrix.

[0065] In this embodiment, the first matrix can be a pre-set key matrix, and the receiver stores a third matrix identical to the first matrix. The first matrix can be set locally on the sender's end, and the third matrix can be set locally on the receiver's end.

[0066] The parameters of the first model can be adjusted using the target dataset to determine the model parameters corresponding to the target dataset. The second matrix is ​​determined using the model parameters corresponding to the target dataset and the first matrix. The second matrix represents the encrypted target dataset. The second matrix can then be transmitted to the receiver, allowing the receiver to obtain the model parameters corresponding to the target dataset by multiplying the third matrix with the second matrix, thereby realizing the transmission of model parameters to the receiver.

[0067] By determining the second matrix, only the second matrix can be transmitted to the receiver. Compared to directly transmitting model parameters, this reduces the amount of data transmitted and improves data transmission efficiency. Furthermore, since the third and second matrices need to be used simultaneously to determine the model parameters corresponding to the target dataset, encrypted data transmission is achieved, improving the security of data transmission.

[0068] This application also proposes a dialogue response method, such as... Figure 4 As shown, it includes the following steps: Step S301: Obtain the problem information sent by the receiver's identifier.

[0069] In this embodiment, the entity executing the dialogue response method can be the receiver or a server between the receiver and the sender. The receiver and sender can engage in dialogue; the sender and receiver can be different electronic devices, each corresponding to a different user. The receiver's identifier can be the identifier of the corresponding electronic device or the identifier of the corresponding user; different receivers correspond to different identifiers. For example, the receiver's identifier can be any of the following forms: device identification code of the electronic device, user ID, user nickname, etc.

[0070] The sender can ask a question to the receiver's identifier, and the receiver's identifier receives the question information. The sender obtains the question information sent by the receiver's identifier.

[0071] Step S302: Determine the response method for the problem information.

[0072] The response method for responding to problem information can be pre-configured, and this response method corresponds to the identifier of the recipient. Different identifiers of recipients can correspond to different response methods.

[0073] Step S303: In response to the response method, model parameters are sent to the identifier of the receiver, the model parameters being loaded into a second model that the receiver can enable so that the second model can answer the question for the sender.

[0074] In this embodiment, the sender can enable the first model to generate model parameters corresponding to the target data set based on the sender's target data set. For example, the sender adjusts the parameters of the first model using the target data set, and determines the adjusted parameters in the first model as the model parameters corresponding to the target data set. In this response method, the sender sends the identifier of the model parameters to the receiver. Subsequently, the receiver can load the model parameters into the second model, enabling the second model to possess knowledge corresponding to the target data set. After inputting the question information into the second model, the second model can accurately infer the response information corresponding to the question information.

[0075] For example, if the target dataset consists of 1000 year-end reports, the sender uses these 1000 reports to train a first model, determining the model parameters corresponding to each of the 1000 reports. These model parameters are then sent to the receiver's identifier. The receiver can subsequently load these parameters into a second model, enabling the second model to possess the knowledge corresponding to the 1000 year-end reports. If the 1000 year-end reports include a target year-end report, and the question is "What does the summary of the target year-end report include?", inputting this question into the second model allows it to accurately infer the summary of the target year-end report.

[0076] The dialogue response method of this application embodiment obtains the question information sent by the identifier of the receiver; determines the response method for the question information; and in response to the response method, sends model parameters to the identifier of the receiver. The model parameters are used to load a second model that the receiver can enable so that the second model can answer the question for the sender. The model parameters are generated by the sender based on the sender's target data set by the first model that can be enabled. By configuring the response method in conjunction with the identifier of the receiver, the receiver can accurately answer the sender's question using the second model, improving the dialogue efficiency between the sender and receivers corresponding to different identifiers. Since the second model has knowledge of the target data set by loading the model parameters, it can transmit the target data set based on the model parameters corresponding to the target data set without complex key management, thereby improving the security of data transmission while reducing computational overhead.

[0077] In some embodiments of this application, the process by which the sender enables a first model to generate model parameters corresponding to the target data set based on the sender's target data set includes: obtaining a first matrix, which is a key matrix; obtaining a second matrix based on the first model, using the target data set and the first matrix, whereby the second matrix represents the encrypted target data set; wherein the second matrix is ​​used to transmit to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by a third matrix, and the first matrix and the third matrix are identical. Sending the identifier of the model parameters to the receiver includes: sending the identifier of the second matrix to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by the third matrix.

[0078] In this embodiment, the first matrix can be a pre-set key matrix, and the receiver stores a third matrix identical to the first matrix. The first matrix can be set locally on the sender's end, and the third matrix can be set locally on the receiver's end.

[0079] The parameters of the first model can be adjusted using the target dataset to determine the model parameters corresponding to the target dataset. The second matrix is ​​determined using the model parameters corresponding to the target dataset and the first matrix. The second matrix represents the encrypted target dataset. The second matrix can then be transmitted to the receiver, allowing the receiver to obtain the model parameters corresponding to the target dataset by multiplying the third matrix with the second matrix, thereby realizing the transmission of model parameters to the receiver.

[0080] By determining the second matrix, only the second matrix can be transmitted to the receiver. Compared to directly transmitting model parameters, this reduces the amount of data transmitted and improves data transmission efficiency. Furthermore, since the third and second matrices need to be used simultaneously to determine the model parameters corresponding to the target dataset, encrypted data transmission is achieved, improving the security of data transmission.

[0081] In some embodiments of this application, determining the response method to the problem information includes: Switching from the first response method to the second response method The first response method is to obtain the user's input information regarding the question information as the response information to the question information; The second response method is to send model parameters to the identifier of the receiver. The model parameters are used to load a second model that the receiver can enable so that the second model can answer the question information for the sender.

[0082] In this embodiment, the response methods for responding to question information include a first response method and a second response method. After obtaining the question information, the system automatically switches from the first response method to the second response method. In the first response method, the user's input information regarding the question information is obtained and used as the response information. In the second response method, the identifier of the model parameters is sent to the receiver. Subsequently, the receiver can load the model parameters into the second model, enabling the second model to possess knowledge corresponding to the target dataset. After the question information is input into the second model, the second model can accurately infer the response information corresponding to the question information.

[0083] By switching from the first response method to the second response method, the system automatically switches from manual replies by the user to automatic replies by the second model, improving dialogue efficiency and enhancing the user experience.

[0084] In some embodiments of this application, the problem information may include prompt words or instruction information. For example, the problem information may be "Please provide a summary of the target year-end report".

[0085] This application also proposes a dialogue response method, such as... Figure 5 As shown, it includes the following steps: Step S401: Obtain the problem information sent by the receiver's identifier.

[0086] In this embodiment, the entity executing the dialogue response method can be the receiver or a server between the receiver and the sender. The receiver and sender can engage in dialogue; the sender and receiver can be different electronic devices, each corresponding to a different user. The receiver's identifier can be the identifier of the corresponding electronic device or the identifier of the corresponding user; different receivers correspond to different identifiers. For example, the receiver's identifier can be any of the following forms: device identification code of the electronic device, user ID, user nickname, etc.

[0087] The sender can ask a question to the receiver's identifier, and the receiver's identifier receives the question information. The sender obtains the question information sent by the receiver's identifier.

[0088] Step S402: Display the problem information in a chat window based on the identifier of the recipient.

[0089] In this embodiment, the chat window of the receiver's identifier can display the dialogue between the sender and the receiver's identifier. After obtaining the problem information, the problem information is displayed in the chat window so that the user can see the problem information intuitively.

[0090] Step S403: If the problem information meets the target conditions, the smart assistant replies with the problem information.

[0091] In this embodiment, the intelligent assistant is an intelligent assistant based on the first model. The first model can be set on the sender and receiver, or on the sender and server. For example, the first model can be a base large language model.

[0092] Determine if the question information meets the target conditions; if so, automatically reply with the question information through the intelligent assistant.

[0093] The sender can enable the first model to generate model parameters corresponding to the sender's target data set. For example, the sender can adjust the parameters of the first model using the target data set, and determine the adjusted parameters in the first model as the model parameters corresponding to the target data set. After loading the model parameters into the receiver's first model, the receiver's first model possesses the knowledge corresponding to the target data set. After inputting the question information into the receiver's first model, the receiver's first model can accurately infer the response information corresponding to the question information, thereby enabling the intelligent assistant to accurately respond to the question information.

[0094] In some implementations of this application, the target conditions may include at least one of the following: the problem information is related to the target data set; the problem information is not sensitive information; the problem information is not a matter of subjective value judgment; the problem information does not violate laws, regulations, or public order and good morals.

[0095] In this embodiment, since the first model possesses knowledge of the target data set, the accuracy of the response information can be improved by relating the question information to the target data set. By ensuring that the question information does not belong to sensitive target information (such as personal ID numbers, mobile phone numbers, bank account information, etc.), the leakage of sensitive target information can be avoided, improving data security. Since subjective value judgments vary from person to person, the accuracy of the response information can be improved by ensuring that the question information does not fall under subjective value judgment questions (such as "Which lifestyle is better?"). Furthermore, by ensuring that the question information does not violate laws, regulations, or social norms (such as racial discrimination, pornography and violence, intellectual property infringement, etc.), illegal responses can be avoided.

[0096] In some embodiments of this application, if the problem information does not meet the target conditions, the user is prompted to manually input the information, and the user's input information is used as the response information for the problem information.

[0097] The dialogue response method of this application embodiment obtains question information sent by the receiver's identifier; displays the question information in a chat window based on the receiver's identifier; and if the question information meets the target conditions, a smart assistant replies to the question information. The smart assistant's reply to the question information is generated by loading model parameters into a first model. The model parameters are those parameters that the sender can enable the first model to generate, based on the sender's target data set, corresponding to the target data set. This scheme achieves automatic reply to the question information through a smart assistant when the question information meets the target conditions, improving the user experience. It also improves the dialogue efficiency between the sender and receivers corresponding to different identifiers. Since the first model possesses knowledge of the target data set by loading model parameters, it can transmit the target data set based on the model parameters corresponding to the target data set without complex key management, thereby improving data transmission security while reducing computational overhead.

[0098] In some embodiments of this application, the process by which the sender enables a first model to generate model parameters corresponding to the target data set based on the sender's target data set includes: obtaining a first matrix, which is a key matrix; obtaining a second matrix based on the first model, using the target data set and the first matrix, whereby the second matrix represents the encrypted target data set; wherein the second matrix is ​​used to transmit to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by a third matrix, and the first matrix and the third matrix are identical. Sending the identifier of the model parameters to the receiver includes: sending the identifier of the second matrix to the receiver so that the receiver obtains the model parameters corresponding to the target data set by multiplying the second matrix by the third matrix.

[0099] In this embodiment, the first matrix can be a pre-set key matrix, and the receiver stores a third matrix identical to the first matrix. The first matrix can be set locally on the sender's end, and the third matrix can be set locally on the receiver's end.

[0100] The parameters of the first model can be adjusted using the target dataset to determine the model parameters corresponding to the target dataset. The second matrix is ​​determined using the model parameters corresponding to the target dataset and the first matrix. The second matrix represents the encrypted target dataset. The second matrix can then be transmitted to the receiver, allowing the receiver to obtain the model parameters corresponding to the target dataset by multiplying the third matrix with the second matrix, thereby realizing the transmission of model parameters to the receiver.

[0101] By determining the second matrix, only the second matrix can be transmitted to the receiver. Compared to directly transmitting model parameters, this reduces the amount of data transmitted and improves data transmission efficiency. Furthermore, since the third and second matrices need to be used simultaneously to determine the model parameters corresponding to the target dataset, encrypted data transmission is achieved, improving the security of data transmission.

[0102] This application also provides a data transmission device, such as... Figure 6 As shown, it includes: a first obtaining module for obtaining transmission data from the sender; a second obtaining module for obtaining model parameters corresponding to the transmission data based on the transmission data and a first model; and a transmission module for transmitting the model parameters to the receiver, wherein the model parameters are loaded into a second model of the receiver so that the second model can answer questions about the transmission data; wherein the first model and the second model are the same model.

[0103] The data transmission device of this application embodiment uses a second obtaining module to convert the transmitted data into model parameters through a first model, and uses a transmission module to transmit the model parameters to the receiver. The receiver can load the model parameters into a second model, so that the second model has the ability to reconstruct the transmitted data. Since the transmitted data is not directly transmitted and complex key management is not required, the device improves the security of data transmission while reducing computational overhead.

[0104] In a specific application scenario, the second obtaining module is specifically used for: obtaining a first matrix, the first matrix being a key matrix; based on the first model, obtaining a second matrix using the transmitted data and the first matrix, the second matrix representing the encrypted transmitted data; wherein, the second matrix is ​​used to transmit to the receiver so that the receiver obtains model parameters corresponding to the transmitted data by multiplying the second matrix by the third matrix, and the first matrix is ​​the same as the third matrix.

[0105] In a specific application scenario, the first matrix is ​​one of the two matrices corresponding to the incremental model parameters obtained by training the first model based on the target file set with the weight parameters frozen; wherein, during each round of data transmission, the parameters of the first matrix remain unchanged; the transmitted data does not belong to the target file set.

[0106] In specific application scenarios, different sets of target files correspond to different first matrices.

[0107] In specific application scenarios, an update module is also included, which is used to update the first matrix according to the change in the number of files in the target file set.

[0108] The electronic device in this application embodiment can be a terminal, or it can be any other device besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. The embodiments disclosed in this disclosure do not impose specific limitations.

[0109] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0110] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A data transmission method, comprising: Obtain the transmitted data from the sender; Based on the transmitted data and the first model, obtain the model parameters corresponding to the transmitted data; The model parameters are transmitted to the receiver, and the model parameters are used to load a second model into the receiver so that the second model can answer questions about the transmitted data. The first model and the second model are the same model.

2. The data transmission method as described in claim 1, wherein obtaining the model parameters corresponding to the transmission data based on the transmission data and the first model includes: Obtain the first matrix, which is the key matrix; Based on the first model, a second matrix is ​​obtained using the transmitted data and the first matrix, and the second matrix represents the encrypted transmitted data. The second matrix is ​​used to transmit to the receiver so that the receiver obtains the model parameters corresponding to the transmitted data by multiplying the third matrix and the second matrix. The first matrix is ​​the same as the third matrix.

3. The data transmission method as described in claim 2, wherein the first matrix is ​​one of the two matrices corresponding to the incremental model parameters obtained by training the first model based on the target file set with frozen weight parameters; in, During each round of data transmission, the parameters of the first matrix remain unchanged; The transmitted data does not belong to the target document set.

4. In the data transmission method as described in claim 3, different target file sets correspond to different first matrices.

5. The data transmission method as described in claim 3, further comprising: The first matrix is ​​updated based on the change in the number of files in the target file set.

6. A dialogue response method, comprising: Obtain the recipient's identifier; Configure the response method for responding to the receiver's identifier dialogue request based on the receiver's identifier; The response method is to send model parameters to the identifier of the receiver. The model parameters are used to load a second model that the receiver can enable so that the second model can answer the questions asked by the sender. The model parameters are the parameters that enable the sender to generate a first model based on the sender's target data set, corresponding to the target data set.

7. A dialogue response method, comprising: Obtain the recipient's identifier and send the problem information; Determine the response method for the aforementioned problem information; In response to the response method, model parameters are sent to the identifier of the receiver, the model parameters being loaded into a second model that the receiver can enable so that the second model can answer the question for the sender; The model parameters are the parameters that enable the sender to generate a first model based on the sender's target data set, corresponding to the target data set.

8. The dialogue response method as described in claim 7, wherein determining the response method for responding to the question information includes: Switching from the first response method to the second response method The first response method is to obtain the user's input information regarding the question information as the response information to the question information; The second response method is to send model parameters to the identifier of the receiver. The model parameters are used to load a second model that the receiver can enable so that the second model can answer the question information for the sender.

9. A dialogue response method, comprising: Obtain the recipient's identifier and send the problem information; The question information is displayed in a chat window based on the identifier of the recipient; If the question information meets the target conditions, the intelligent assistant will respond with the question information; The intelligent assistant's response to the question is generated by loading model parameters through the first model to respond to the question. The model parameters are the parameters that enable the sender to generate a first model based on the sender's target data set, corresponding to the target data set.

10. A data transmission apparatus, comprising: The first acquisition module is used to acquire the transmitted data from the sender; The second obtaining module is used to obtain model parameters corresponding to the transmission data based on the transmission data and the first model; A transmission module is used to transmit the model parameters to a receiver, the model parameters being loaded into a second model at the receiver so that the second model can answer questions about the transmitted data; The first model and the second model are the same model.

Images