Method, device and storage medium for running evaluation of privacy computing algorithm

Abstract

The one or more embodiments of the application provide a method, device and storage medium for evaluating running of a privacy computing algorithm, the method comprising: obtaining algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device for running the privacy computing algorithm; inputting the algorithm running information and the device configuration information into a service model of a large language model, to predict, by the service model of the large language model, a running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; the service model of the large language model being a model obtained by performing supervised fine-tuning training on a base model of the large language model based on training samples including the algorithm running information and the device configuration information and labeled with the running condition; obtaining the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputting the running condition.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present application relate to the technical field of artificial intelligence, and in particular to a method and device for evaluating the running of a privacy computing algorithm, and a storage medium. BACKGROUND

[0002] Privacy computing is a technology for protecting data privacy, and its purpose is to protect data privacy during computation. It can perform data analysis and computation without exposing raw data. Privacy computing generally includes data encryption, privacy protection protocols, anonymization techniques, secure multi-party computation, etc. Among them, data encryption refers to encrypting data using various encryption algorithms. Privacy protection protocols refer to ensuring the privacy and security of data through differential privacy, homomorphic encryption, etc. Anonymization techniques refer to separating data from personal identity through anonymization processing. Secure multi-party computation refers to computing by multiple data providers without disclosing their private data.

[0003] In the scenario of privacy computing, multi-party cooperation is often faced. In the privacy computing algorithm realized through multi-party cooperation, the original data does not leave the domain, but is encrypted in various ways, and is calculated based on the ciphertext data, while various protocols are used, and is interacted based on the ciphertext data, etc. These data processing processes are black boxes, that is, the input and output are known, but the internal running logic is invisible to the user, so whether the privacy computing algorithm runs in the agreed manner, whether the data is maliciously increased in a large scale in the running process of the privacy computing algorithm to collide with the data of the cooperation party, these problems are usually difficult to analyze and judge. In this case, how to evaluate the running of the privacy computing algorithm has become a problem to be solved. SUMMARY

[0004] One or more embodiments of the present application provide technical solutions as follows:

[0005] The present application provides a model construction method based on pre-training, which comprises:

[0006] Constructing a training sample; wherein the training sample includes algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device for running the privacy computing algorithm; the training sample is labeled with the running condition of the privacy computing algorithm on the device;

[0007] The base model of the large language model is supervised fine-tuning trained based on the training sample, to obtain a service model of the large language model; wherein the service model of the large language model is used to predict the running situation of the privacy computing algorithm on the device based on algorithm running information corresponding to the privacy computing algorithm to be evaluated and device configuration information of the device used to run the privacy computing algorithm.

[0008] The present application also provides a method for evaluating the running of a privacy computing algorithm, comprising:

[0009] obtaining algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device used to run the privacy computing algorithm;

[0010] inputting the algorithm running information and the device configuration information into a service model of a large language model, so that the service model of the large language model predicts the running situation of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; wherein the service model of the large language model is a model obtained by supervised fine-tuning training of a base model of the large language model based on a training sample; the training sample comprises algorithm running information corresponding to a preset privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm; the training sample is labeled with the running situation of the privacy computing algorithm on the device;

[0011] obtaining the running situation of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputting the running situation.

[0012] The present application also provides a pre-training-based model construction device, comprising:

[0013] a construction module for constructing a training sample; wherein the training sample comprises algorithm running information corresponding to a preset privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm; the training sample is labeled with the running situation of the privacy computing algorithm on the device;

[0014] a training module for supervised fine-tuning training of a base model of a large language model based on the training sample, to obtain a service model of the large language model; wherein the service model of the large language model is used to predict the running situation of the privacy computing algorithm on the device based on algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device used to run the privacy computing algorithm.

[0015] The present application also provides a device for evaluating the running of a privacy computing algorithm, comprising:

[0016] an acquisition module, configured to acquire algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device used to run the privacy computing algorithm;

[0017] a prediction module, configured to input the algorithm running information and the device configuration information into a service model of a large language model to predict, by the service model of the large language model, a running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; the service model of the large language model is a model obtained by supervised fine-tuning training of a base model of the large language model based on training samples; the training samples include algorithm running information corresponding to a preset privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm; the training samples are labeled with a running condition of the privacy computing algorithm on the device;

[0018] an output module, configured to acquire the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model and output the running condition.

[0019] The application further provides an electronic device, comprising:

[0020] a processor;

[0021] a memory for storing processor-executable instructions;

[0022] The processor implements the steps of the method according to any one of the above by running the executable instructions.

[0023] The application further provides a computer-readable storage medium having computer instructions stored thereon, the instructions being executed by a processor to implement the steps of the method according to any one of the above.

[0024] In the above technical solution, the service model of the large language model for running evaluation of the privacy computing algorithm obtained by fine-tuning training can be used to input the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm into the service model of the large language model, so that the service model of the large language model predicts the running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information, and outputs the running condition predicted by the service model of the large language model. In this way, since the service model of the large language model for running evaluation of the privacy computing algorithm obtained by fine-tuning training can be directly used to evaluate the running of the privacy computing algorithm to be evaluated, the difficulty of evaluating the running of the privacy computing algorithm can be reduced, and the accuracy and diversity of the result of the running evaluation of the privacy computing algorithm can be ensured. BRIEF DESCRIPTION OF DRAWINGS

[0025] The drawings needed to be used in the following description of the exemplary embodiments will be described below, wherein:

[0026] Figure 1 is a flow chart of a pre-training based model construction method according to an exemplary embodiment of the present application.

[0027] Figure 2 is a flow chart of a running evaluation method of a privacy computing algorithm according to an exemplary embodiment of the present application.

[0028] Figure 3 is a structural schematic diagram of a device according to an exemplary embodiment of the present application.

[0029] Figure 4 is a block diagram of a pre-training based model construction apparatus according to an exemplary embodiment of the present application.

[0030] Figure 5 is a block diagram of a running evaluation apparatus of a privacy computing algorithm according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0031] The exemplary embodiments will be described in detail herein with reference to the drawings. The following description is with reference to the drawings, wherein like numerals indicate like elements throughout the several figures. The implementation described in the following exemplary embodiments does not represent all implementations consistent with one or more embodiments of the present application. Instead, they are merely examples in accordance with some aspects of one or more embodiments of the present application.

[0032] It should be noted that the steps of the methods in accordance with the present application need not necessarily be performed in the order shown and described herein. In some other embodiments, the methods can include more or fewer steps than those described herein. Moreover, the individual steps described herein can in some other embodiments be broken down into multiple steps for purposes of description, and multiple steps described herein can in some other embodiments be combined into a single step for purposes of description.

[0033] In the scenario of privacy computing, multi-party cooperation is often faced. In the privacy computing algorithm realized through multi-party cooperation, the original data does not leave the domain, but various encryptions are performed on the original data, and calculations are performed based on the ciphertext data, various protocols are used, and interactions are performed based on the ciphertext data, etc. These data processing processes are black boxes, that is, the input and output are known, but the internal running logic is invisible to the user, so it is difficult to analyze and judge whether the privacy computing algorithm is running in the agreed manner, and whether the data is maliciously increased in a large scale to collide with the data of the cooperation party during the running of the privacy computing algorithm. Therefore, it is difficult to evaluate the running of the privacy computing algorithm.

[0034] In addition, the scenario of privacy computing is diverse, and the implementation of the privacy computing algorithm has strong diversity, which becomes extremely complex when combined, resulting in multiple possibilities of the internal running logic of the privacy computing algorithm. Taking a private set intersection (PSI) algorithm as an example, there are hundreds of possibilities for its internal running logic. In addition to the changes in data size and hardware environment, it becomes more difficult to evaluate the running of the privacy computing algorithm.

[0035] The present application provides a technical solution for evaluating the running of a privacy computing algorithm. Specifically, the technical solution provided by the present application uses a large language model (LLM) to complete the evaluation of the running of the privacy computing algorithm.

[0036] A large language model refers to a deep learning model trained using a large amount of text data, which can be used to generate natural language text or understand the meaning of language text. A large language model can handle a variety of natural language tasks, such as text classification, question answering, dialogue, etc., and is an important way to artificial intelligence.

[0037] A large language model usually adopts a Transformer architecture, that is, a large language model is usually a deep learning model based on a Transformer architecture. A deep learning model based on a Transformer architecture is a type of neural network model that adopts a Transformer architecture, which performs well in natural language processing and other fields.

[0038] Transformer is a neural network model for sequence-to-sequence (Sequence-to-Sequence) modeling. Transformer does not need to rely on a recursive structure, and can parallelize training and inference, speeding up model processing.

[0039] In deep learning models based on the Transformer architecture, a multi-layer Transformer encoder is usually used to extract features from the input sequence, and a Transformer decoder is used to convert these features into an output sequence. At the same time, these models also usually use a self-attention mechanism to capture long-distance dependencies in the input sequence, and a residual connection and a normalization method to speed up training and improve model performance.

[0040] A pre-trained model is a language model pre-trained on large-scale unlabeled text data. The pre-trained model is general and is not designed and optimized for a specific task. In order to adapt the pre-trained model to a specific task, a fine-tuning operation needs to be performed to improve the performance of the model on the specific task. A large language model is a model that is further fine-tuned based on a pre-trained model and performs supervised learning through labeled text data.

[0041] That is, the training process of a large language model can generally be divided into two stages: pre-training and fine-tuning. In the pre-training stage, unsupervised learning (e.g., self-supervised learning) can be used to pre-train on public datasets (e.g., web encyclopedias, web articles, books, etc.). The model will try to predict missing parts or the next word based on the context to learn the internal structure of the language and the internal rules of the language. Gradient descent and other optimization algorithms can be used to continuously adjust the model parameters to gradually improve the performance of the model on the pre-training task. In the fine-tuning stage, according to the specific application scenario and task requirements, a corresponding supervised learning task (e.g., text classification, named entity recognition, question and answer system, etc.) can be selected and a task-specific dataset can be prepared. The pre-trained model is used as the starting point for fine-tuning, and supervised learning is used to train on the task-specific dataset. Gradient descent and other optimization algorithms can be used to adjust the model parameters based on the labels of the dataset, so that the model gradually adapts to the requirements of the specific task and dataset. Pre-training and fine-tuning are complementary processes. Pre-training enables the model to have broad language understanding capabilities, and fine-tuning makes the model more specialized and accurate on specific tasks.

[0042] Generally, the pre-trained large language model is referred to as the base model of the large language model, and the fine-tuned large language model is referred to as the service model of the large language.

[0043] In the technical solutions provided in the present application, in order to use the large language model to complete the running evaluation of the privacy computing algorithm, the supervised learning manner can be adopted to fine-tune the training on the task-specific data set of the running evaluation of the privacy computing algorithm, so as to obtain the service model of the large language model that can be used for the running evaluation of the privacy computing algorithm. In this way, the service model of the large language model can be used to complete the running evaluation of the privacy computing algorithm.

[0044] The technical solutions provided in the present application will be described in detail from the following two aspects: obtaining the service model of the large language model for the running evaluation of the privacy computing algorithm, and using the service model of the large language model to complete the running evaluation of the privacy computing algorithm.

[0045] Please refer to Figure 1 , Figure 1 is a flowchart of a pre-training-based model construction method according to an example embodiment of the present application.

[0046] In the present embodiment, the base model of the large language model constructed by pre-training can be directly used, and the construction of the service model of the large language model for the running evaluation of the privacy computing algorithm can be further completed by fine-tuning the base model of the large language model.

[0047] Specifically, the running evaluation task of the privacy computing algorithm can be taken as a supervised learning task during fine-tuning, and a task-specific data set for the running evaluation of the privacy computing algorithm can be prepared; the base model of the large language model can be taken as the starting point of fine-tuning, and the supervised learning manner can be adopted to train on the task-specific data set for the running evaluation of the privacy computing algorithm; the model parameters can be adjusted according to the labels of the data set by using the gradient descent and other optimization algorithms, so that the model gradually adapts to the requirements of the specific task and data set, thereby obtaining the service model of the large language model for the running evaluation of the privacy computing algorithm.

[0048] In some embodiments, the pre-training-based model construction method described above can be applied to a server. The server can be a server including a single independent physical host, or a server cluster composed of multiple independent physical hosts; or the server can be a virtual server, a cloud server, etc. carried by a host cluster. Alternatively, the pre-training-based model construction method described above can be applied to desktop computers, notebook computers, personal digital assistants (PDAs), tablet devices, and other electronic devices with certain computing capabilities.

[0049] As shown in Figure 1 , the pre-training-based model construction method described above can include the following steps:

[0050] Step 102: constructing a training sample; wherein the training sample includes algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device used for running the privacy computing algorithm; and the training sample is labeled with a running condition of the privacy computing algorithm on the device.

[0051] In this embodiment, a training sample specific to the running evaluation task of the privacy computing algorithm can be constructed as the prepared data set specific to the running evaluation task of the privacy computing algorithm.

[0052] For any one of the constructed training samples, the training sample can include algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device used for running the privacy computing algorithm; and the training sample can be labeled with a running condition of the privacy computing algorithm on the device as a label of the training sample.

[0053] Step 104: based on the training sample, performing supervised fine-tuning training on a base model of a large language model to obtain a service model of the large language model; wherein the service model of the large language model is used to predict a running condition of a privacy computing algorithm corresponding to algorithm running information to be evaluated and device configuration information of a device used for running the privacy computing algorithm.

[0054] In this embodiment, after the above-mentioned training sample is constructed, the base model of the large language model constructed through pre-training can be further subjected to supervised fine-tuning training based on the training sample to obtain a service model of the large language model for running evaluation of the privacy computing algorithm.

[0055] Because the content of the above-mentioned training sample is algorithm running information corresponding to a privacy computing algorithm, and device configuration information of a device used for running the privacy computing algorithm, and the label is a running condition of the privacy computing algorithm on the device, the above-mentioned service model of the large language model can be specifically used to predict a running condition of a privacy computing algorithm corresponding to algorithm running information to be evaluated and device configuration information of a device used for running the privacy computing algorithm. That is, when the service model of the large language model is used subsequently, algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device used for running the privacy computing algorithm can be used as input, and the output is the predicted running condition of the privacy computing algorithm on the device.

[0056] In some embodiments, the algorithm running information can include a combination of one or more of the following: an algorithm protocol, an algorithm description, a data size of data used for calculation during algorithm running. For example, the algorithm protocol can be ECDH (Elliptic Curve Diffie-Hellman) FourQ curve; the algorithm description can be: the data size of the first encryption result is 32 bytes, the second encryption result is truncated into a data segment with a data size of 10 bytes, the party with the result needs to obtain the second ciphertext of both parties and perform intersection calculation, the CPU performance during the first encryption and the second encryption is 50,000 calculation operations per second, A and B can perform parallel calculation, and A takes the result; the data size of the data used for calculation during algorithm running can be 100 million data records, or can be 100 million rows of 1 column data, and the data size of each row of data is 64 bytes.

[0057] The device configuration information can include a combination of one or more of the following: device computing power, data transmission capability of the device. The device computing power can be the CPU frequency and the number of CPU cores of the device; the data transmission capability of the device can be the bandwidth of the device.

[0058] The running condition can include the time consumption of algorithm running.

[0059] In this case, the service model of the large language model can predict the time consumption of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm.

[0060] It should be noted that if the privacy computing algorithm includes multiple stages (for example: a first encryption stage, a second encryption stage, etc.), the service model of the large language model can predict the time consumption of each stage of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm.

[0061] In some embodiments, the algorithm running information can include a combination of one or more of the following: an algorithm protocol, an algorithm description, a data size of data used for calculation during algorithm running, and a running log generated during algorithm running.

[0062] The device configuration information can include a combination of one or more of the following: device computing power, data transmission capability of the device.

[0063] The running condition can include a diagnosis result indicating whether the calculation behavior during algorithm running meets the expectation.

[0064] In this case, the service model of the large language model can predict whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm, and output a diagnosis result indicating whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation.

[0065] Specifically, in some embodiments, the service model of the large language model can predict the time consumption of the privacy computing algorithm running on the device based on the algorithm protocol, algorithm description and data scale of the data used for computing in the algorithm running process corresponding to the privacy computing algorithm to be evaluated, and the device configuration information of the device used to run the privacy computing algorithm.

[0066] Further, the service model of the large language model can compare the algorithm running time recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted time consumption of the privacy computing algorithm running on the device to determine whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation, and generate a diagnosis result indicating whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation.

[0067] For example, if a certain privacy computing algorithm is not suitable for using cache, but cache is used in the running process of the privacy computing algorithm, it will cause the CPU computing time consumption to be greatly reduced. In this way, if the algorithm running time recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm is much less than the predicted time consumption of the privacy computing algorithm running on the device, it can be determined that the computing behavior in the running process of the privacy computing algorithm does not conform to the expectation, and thus a diagnosis result indicating that the computing behavior in the running process of the privacy computing algorithm does not conform to the expectation can be generated.

[0068] If the privacy computing algorithm contains multiple stages, the service model of the large language model can predict the time consumption of each stage of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm. In this case, the service model of the large language model can compare the algorithm running time of each stage recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted time consumption of each stage of the privacy computing algorithm running on the device to determine whether the computing behavior of each stage in the running process of the privacy computing algorithm conforms to the expectation, and generate a diagnosis result indicating whether the computing behavior of each stage in the running process of the privacy computing algorithm conforms to the expectation.

[0069] In some embodiments, the service model of the large language model can predict the traffic of the privacy computing algorithm when running on the device based on the algorithm protocol, the algorithm description, and the data scale of the data used for calculation in the algorithm running process corresponding to the privacy computing algorithm to be evaluated, and the device configuration information of the device used for running the privacy computing algorithm.

[0070] Further, the service model of the large language model can compare the traffic recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted traffic of the privacy computing algorithm when running on the device to determine whether the calculation behavior in the algorithm running process of the privacy computing algorithm is as expected, and generate a diagnosis result indicating whether the calculation behavior in the algorithm running process of the privacy computing algorithm is as expected.

[0071] For example, if the security strength requirement of a certain privacy computing algorithm requires the data size of the ciphertext data to be at least 10 bytes, i.e., the predicted traffic of the privacy computing algorithm when running on the device is not less than 10 bytes, but the traffic recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm is less than 10 bytes, it can be determined that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected, and thus a diagnosis result indicating that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected can be generated.

[0072] Or, if the traffic recorded in the running log generated in the algorithm running process corresponding to a certain privacy computing algorithm is much larger than the predicted traffic of the privacy computing algorithm when running on the device, it can be determined that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected, and thus a diagnosis result indicating that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected can be generated.

[0073] If the privacy computing algorithm contains multiple stages, the service model of the large language model can predict the traffic of each stage of the privacy computing algorithm when running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used for running the privacy computing algorithm. In this case, the service model of the large language model can compare the traffic of each stage recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted traffic of each stage of the privacy computing algorithm when running on the device to determine whether the calculation behavior of each stage in the algorithm running process of the privacy computing algorithm is as expected, and generate a diagnosis result indicating whether the calculation behavior of each stage in the algorithm running process of the privacy computing algorithm is as expected.

[0074] In some embodiments, the privacy computing algorithm mentioned above can be a secure multi-party computation algorithm for privacy computation based on data in the form of ciphertext held by multiple data providers.

[0075] In the technical solution mentioned above, the running evaluation task of the privacy computing algorithm is taken as a supervised learning task during fine-tuning training, and training samples specific to the running evaluation task of the privacy computing algorithm are constructed, wherein the training samples include algorithm running information corresponding to a preset privacy computing algorithm, device configuration information of a device used to run the privacy computing algorithm, and are labeled with running conditions of the privacy computing algorithm on the device. Based on the constructed training samples, supervised fine-tuning training is performed on the base model of the large language model, and a service model of the large language model for running evaluation of the privacy computing algorithm is obtained, so that the service model of the large language model can be directly used for subsequent evaluation of the running of the privacy computing algorithm.

[0076] Please refer to Figure 2 , Figure 2 is a flowchart of a method for running evaluation of a privacy computing algorithm according to an example embodiment of the present application.

[0077] In this embodiment, the service model of the large language model for running evaluation of the privacy computing algorithm constructed by the embodiment shown in Figure 1 can be directly used for evaluation of the running of the privacy computing algorithm.

[0078] In some embodiments, the pre-trained model construction method mentioned above can be applied to a server. The server can be a server including a single independent physical host, or a server cluster composed of multiple independent physical hosts; or the server can be a virtual server, a cloud server, etc. carried by a host cluster. Alternatively, the pre-trained model construction method mentioned above can be applied to desktop computers, notebook computers, personal digital assistants (PDAs), tablet devices, and other electronic devices with certain computing capabilities.

[0079] As shown in Figure 2 , the pre-trained model construction method mentioned above can include the following steps:

[0080] Step 202: Obtain algorithm running information corresponding to a privacy computing algorithm to be evaluated, and device configuration information of a device used to run the privacy computing algorithm.

[0081] In this embodiment, for a privacy computing algorithm to be evaluated, in order to evaluate the running of the privacy computing algorithm, algorithm running information corresponding to the privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm can be acquired first.

[0082] Step 204: inputting the algorithm running information and the device configuration information into a service model of a large language model, to predict, by the service model of the large language model, a running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; wherein the service model of the large language model is a model obtained by supervised fine-tuning training of a base model of the large language model based on training samples; the training samples include algorithm running information corresponding to a preset privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm; and the training samples are labeled with a running condition of the privacy computing algorithm on the device.

[0083] In this embodiment, the algorithm running information corresponding to the above-mentioned privacy computing algorithm and the device configuration information of the device used to run the privacy computing algorithm can be input into the above-mentioned service model of the large language model, to predict, by the service model of the large language model, a running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information.

[0084] Wherein, the process of obtaining the service model of the large language model can refer to the service model of the large language model, which will not be described in detail in this embodiment.

[0085] Step 206: acquiring the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputting the running condition.

[0086] In this embodiment, the output of the above-mentioned service model of the large language model is the running condition of the above-mentioned privacy computing algorithm on the above-mentioned device predicted by the service model of the large language model. In this case, the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model can be acquired, and the acquired running condition can be output. For example, the acquired running condition can be output through a user-interactive interface for the user to view.

[0087] In some embodiments, the algorithm running information can include a combination of one or more of the following: an algorithm protocol, an algorithm description, a data size of data used for calculation during algorithm running. For example, the algorithm protocol can be ECDH (Elliptic Curve Diffie-Hellman) FourQ curve; the algorithm description can be: the data size of the first encryption result is 32 bytes, the second encryption result is truncated into a data segment with a data size of 10 bytes, the party with the result needs to obtain the second ciphertext of both parties and perform intersection calculation, the CPU performance during the first encryption and the second encryption is 50,000 calculation operations per second, A and B can perform parallel calculation, and A takes the result; the data size of the data used for calculation during algorithm running can be 100 million data records, or can be 100 million rows of 1 column data, and the data size of each row of data is 64 bytes.

[0088] The device configuration information can include a combination of one or more of the following: device computing power, data transmission capability of the device. Among them, the device computing power can be the CPU frequency and the number of CPU cores of the device; the data transmission capability of the device can be the bandwidth of the device.

[0089] The running situation can include: algorithm running time consumption.

[0090] In this case, the service model of the large language model can predict the time consumption of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm.

[0091] It should be noted that if the privacy computing algorithm includes multiple stages (for example: a first encryption stage, a second encryption stage, etc.), the service model of the large language model can predict the time consumption of each stage of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm.

[0092] In some embodiments, the algorithm running information can include a combination of one or more of the following: an algorithm protocol, an algorithm description, a data size of data used for calculation during algorithm running, and a running log generated during algorithm running.

[0093] The device configuration information can include a combination of one or more of the following: device computing power, data transmission capability of the device.

[0094] The running situation can include: a diagnosis result indicating whether the calculation behavior during algorithm running meets the expectation.

[0095] In this case, the service model of the large language model can predict whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm, and output a diagnosis result indicating whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation.

[0096] Specifically, in some embodiments, the service model of the large language model can predict the time consumption of the privacy computing algorithm running on the device based on the algorithm protocol, algorithm description and data scale of the data used for computing in the algorithm running process corresponding to the privacy computing algorithm to be evaluated, and the device configuration information of the device used to run the privacy computing algorithm.

[0097] Further, the service model of the large language model can compare the algorithm running time recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted time consumption of the privacy computing algorithm running on the device to determine whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation, and generate a diagnosis result indicating whether the computing behavior in the running process of the privacy computing algorithm conforms to the expectation.

[0098] For example, if a certain privacy computing algorithm is not suitable for using cache, but cache is used in the running process of the privacy computing algorithm, it will cause the CPU computing time consumption to be greatly reduced. In this way, if the algorithm running time recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm is much less than the predicted time consumption of the privacy computing algorithm running on the device, it can be determined that the computing behavior in the running process of the privacy computing algorithm does not conform to the expectation, and thus a diagnosis result indicating that the computing behavior in the running process of the privacy computing algorithm does not conform to the expectation can be generated.

[0099] If the privacy computing algorithm contains multiple stages, the service model of the large language model can predict the time consumption of each stage of the privacy computing algorithm running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm. In this case, the service model of the large language model can compare the algorithm running time of each stage recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted time consumption of each stage of the privacy computing algorithm running on the device to determine whether the computing behavior of each stage in the running process of the privacy computing algorithm conforms to the expectation, and generate a diagnosis result indicating whether the computing behavior of each stage in the running process of the privacy computing algorithm conforms to the expectation.

[0100] In some embodiments, the service model of the large language model can predict the traffic of the privacy computing algorithm when running on the device based on the algorithm protocol, the algorithm description, and the data scale of the data used for calculation in the algorithm running process corresponding to the privacy computing algorithm to be evaluated, and the device configuration information of the device used for running the privacy computing algorithm.

[0101] Further, the service model of the large language model can compare the traffic recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted traffic of the privacy computing algorithm when running on the device, to determine whether the calculation behavior in the algorithm running process of the privacy computing algorithm is as expected, and generate a diagnosis result indicating whether the calculation behavior in the algorithm running process of the privacy computing algorithm is as expected.

[0102] For example, if the security strength requirement of a certain privacy computing algorithm requires the data size of the ciphertext data to be at least 10 bytes, that is, the predicted traffic of the privacy computing algorithm when running on the device is not less than 10 bytes, but the traffic recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm is less than 10 bytes, it can be determined that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected, and thus a diagnosis result indicating that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected can be generated.

[0103] Or, if the traffic recorded in the running log generated in the algorithm running process corresponding to a certain privacy computing algorithm is much larger than the predicted traffic of the privacy computing algorithm when running on the device, it can be determined that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected, and thus a diagnosis result indicating that the calculation behavior in the algorithm running process of the privacy computing algorithm is not as expected can be generated.

[0104] If the privacy computing algorithm contains multiple stages, the service model of the large language model can predict the traffic of each stage of the privacy computing algorithm when running on the device based on the algorithm running information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used for running the privacy computing algorithm. In this case, the service model of the large language model can compare the traffic of each stage recorded in the running log generated in the algorithm running process corresponding to the privacy computing algorithm to be evaluated with the predicted traffic of each stage of the privacy computing algorithm when running on the device, to determine whether the calculation behavior of each stage in the algorithm running process of the privacy computing algorithm is as expected, and generate a diagnosis result indicating whether the calculation behavior of each stage in the algorithm running process of the privacy computing algorithm is as expected.

[0105] In some embodiments, the privacy computing algorithm described above can be a secure multi-party computation algorithm that performs privacy computation based on data in the form of ciphertext held by multiple data providers.

[0106] For example, assume that the data providers of a certain privacy computing algorithm include parties A and B, the number of CPU cores of the computing devices of A and B is 8, the bandwidth of the computing devices of A and B is 100M, the data size of the data provided by A and B for computation during the running of the privacy computing algorithm is 100 million data records, and the algorithm protocol of the privacy computing algorithm is ECDH FourQ curve. The algorithm description of the privacy computing algorithm includes: the data size of the first encryption result is 32 bytes, the second encryption result is truncated into a data segment with a data size of 10 bytes, the party that takes the result needs to obtain the second ciphertext of both parties and perform intersection calculation, the CPU performance during the first encryption and the second encryption is 50,000 calculation operations per second, A and B can perform parallel computation, and A takes the result.

[0107] In this case, the algorithm running time of the privacy computing algorithm predicted by the service model of the large language model described above when running on the device can be determined according to the following steps:

[0108] First, A and B each perform first encryption, which is performed concurrently by 8 cores of the CPU. The time consumption of this stage is: 100 million (data records) ÷ 50,000 (calculation operations) ÷ 8 (core number) = 250 seconds.

[0109] Then, A and B send each other full-amount first ciphertext data. The time consumption of this stage is: 100 million (data records) x 32 (bytes) ÷ 1024 ÷ 1024 ÷ 100 (M bandwidth) x 8 = 244 seconds.

[0110] Then, A and B each perform second encryption on the first ciphertext data of the other party, which is performed concurrently by 8 cores of the CPU. The time consumption of this stage is: 100 million (data records) ÷ 50,000 (calculation operations) ÷ 8 (core number) = 250 seconds.

[0111] Finally, B sends the full-amount second ciphertext data of A to A. The time consumption of this stage is: 100 million (data records) x 10 (bytes) ÷ 1024 ÷ 1024 ÷ 100 (M bandwidth) x 8 = 76 seconds.

[0112] Therefore, the theoretical time of the privacy computing algorithm described above is: 250 seconds + 244 seconds + 250 seconds + 76 seconds = 820 seconds.

[0113] In addition to the empirical values of the time consumption of data reading, intersection calculation, and file writing, the algorithm running time of the privacy computing algorithm described above when running on the device can be completely estimated.

[0114] In the technical solution, the service model of the large language model for running evaluation of a privacy computing algorithm trained through fine-tuning can be used, algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device for running the privacy computing algorithm are input into the service model of the large language model, the running situation of the privacy computing algorithm on the device is predicted based on the algorithm running information and the device configuration information by the service model of the large language model, and the running situation predicted by the service model of the large language model is output. In this way, since the service model of the large language model for running evaluation of a privacy computing algorithm trained through fine-tuning can be directly used to evaluate the running of a privacy computing algorithm to be evaluated, the difficulty of evaluating the running of a privacy computing algorithm can be reduced, and the accuracy and diversity of the result of the running evaluation of the privacy computing algorithm can be ensured.

[0115] Corresponding to the embodiments of the foregoing method, the application also provides embodiments of an apparatus.

[0116] Please refer to Figure 3 , Figure 3 is a structural schematic diagram of a device according to an exemplary embodiment of the application. At the hardware level, the device includes a processor 302, an internal bus 304, a network interface 306, a memory 308, and a non-volatile memory 310, and of course can also include other required hardware. One or more embodiments of the application can be implemented in a software manner, such as reading a corresponding computer program from the non-volatile memory 310 into the memory 308 by the processor 302 and then running. Of course, in addition to the software implementation, one or more embodiments of the application do not exclude other implementation manners, such as a logic device or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logical module, but can also be hardware or a logic device.

[0117] Please refer to Figure 4 , Figure 4 is a block diagram of a model construction apparatus based on pre-training according to an exemplary embodiment of the application.

[0118] The above-mentioned model construction apparatus based on pre-training can be applied to the device shown in Figure 3 to implement the technical solution of the application. The apparatus includes:

[0119] The construction module 402 constructs a training sample; wherein the training sample includes algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device for running the privacy computing algorithm; and the training sample is labeled with a running situation of the privacy computing algorithm on the device;

[0120] The training module 404 performs supervised fine-tuning training on the base model of the large language model based on the training samples to obtain a service model of the large language model; wherein the service model of the large language model is used to predict the running situation of the privacy computing algorithm on the device based on algorithm running information corresponding to the privacy computing algorithm to be evaluated and device configuration information of a device used to run the privacy computing algorithm.

[0121] Optionally, the algorithm running information includes: algorithm protocol, algorithm description, and data scale of data used for calculation during algorithm running; and the device configuration information includes: device computing power and data transmission capability of the device; and the running situation includes: algorithm running time consumption.

[0122] Optionally, the algorithm running information includes: algorithm protocol, algorithm description, data scale of data used for calculation during algorithm running, and running log generated during algorithm running; and the device configuration information includes: device computing power and data transmission capability of the device; and the running situation includes: diagnostic result indicating whether the calculation behavior during algorithm running is in line with expectations.

[0123] Optionally, the privacy computing algorithm is a secure multi-party computation algorithm for performing privacy computation on data in the form of ciphertext held by multiple data providers.

[0124] Please refer to Figure 5 , Figure 5 is a block diagram of a privacy computing algorithm running evaluation device according to an example embodiment of the present application.

[0125] The privacy computing algorithm running evaluation device described above can be applied to the device shown in Figure 3 to implement the technical solution of the present application. The device includes:

[0126] The acquisition module 502 acquires algorithm running information corresponding to a privacy computing algorithm to be evaluated and device configuration information of a device used to run the privacy computing algorithm;

[0127] The prediction module 504 inputs the algorithm running information and the device configuration information into a service model of a large language model, so that the service model of the large language model predicts the running situation of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; wherein the service model of the large language model is a model obtained by performing supervised fine-tuning training on a base model of the large language model based on training samples; the training samples include algorithm running information corresponding to a preset privacy computing algorithm and device configuration information of a device used to run the privacy computing algorithm; and the training samples are labeled with the running situation of the privacy computing algorithm on the device.

[0128] The output module 506 obtains the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputs the running condition.

[0129] Optionally, the algorithm running information includes: algorithm protocol, algorithm description, and data scale of data used for calculation during algorithm running; the device configuration information includes: device computing power and data transmission capability of the device; and the running condition includes: algorithm running time consumption.

[0130] Optionally, the algorithm running information includes: algorithm protocol, algorithm description, data scale of data used for calculation during algorithm running, and running log generated during algorithm running; the device configuration information includes: device computing power and data transmission capability of the device; and the running condition includes: diagnosis result indicating whether the calculation behavior during algorithm running is expected.

[0131] Optionally, the prediction of the running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information includes:

[0132] prediction of algorithm running time consumption of the privacy computing algorithm on the device based on the algorithm protocol, the algorithm description, the data scale of data used for calculation during algorithm running, and the device configuration information;

[0133] comparison of algorithm running time consumption recorded in the running log generated during algorithm running with the predicted algorithm running time consumption, to determine whether the calculation behavior during algorithm running is expected, and generation of a diagnosis result indicating whether the calculation behavior during algorithm running is expected.

[0134] Optionally, the prediction of the running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information includes:

[0135] prediction of traffic when the privacy computing algorithm runs on the device based on the algorithm protocol, the algorithm description, the data scale of data used for calculation during algorithm running, and the device configuration information;

[0136] comparison of traffic recorded in the running log generated during algorithm running with the predicted traffic, to determine whether the calculation behavior during algorithm running is expected, and generation of a diagnosis result indicating whether the calculation behavior during algorithm running is expected.

[0137] Optionally, the privacy computing algorithm is a secure multi-party computing algorithm for privacy computing based on data in the form of ciphertext held by multiple data providers.

[0138] For apparatus embodiments, they basically correspond to the method embodiments, therefore the relevant parts are referred to the parts of the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, i.e., can be located in one place, or can be distributed to multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purposes of the technical solutions of the present application.

[0139] The systems, apparatuses, modules or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, and the specific form of the computer can be 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 e-mail device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0140] In a typical configuration, a computer includes one or more processors (CPUs), input / output interfaces, network interfaces, and memories.

[0141] The memory can include non-persistent memory in computer readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or Flash memory. The memory is an example of computer readable media.

[0142] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented by any method or technology for storing 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 technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, disk storage, quantum memory, graphene-based storage medium or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition in this paper, computer readable media does not include transitory computer readable media, such as modulated data signals and carriers.

[0143] It is to be understood that the terms "including", "comprising", or any other variation thereof, are intended to cover the contents "non-exclusive inclusion" so that a process, method, article, or apparatus that includes a list of elements does not only include those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0144] The above describes specific embodiments of the present application. Other embodiments are within the scope of the present application. In some cases, the actions or steps described in the present application can be performed in an order different from the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0145] The terms used in one or more embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit one or more embodiments of the present application. The singular forms "a", "said" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. The term "and / or" means and includes any or all possible combinations of one or more associated listed items.

[0146] The description of the terms "one embodiment", "some embodiments", "example", "specific example" or "one implementation" and the like used in one or more embodiments of the present application means that the specific features or characteristics described in connection with the embodiment are included in at least one embodiment of the present application. The illustrative description of these terms does not necessarily refer to the same embodiment. Moreover, the specific features or characteristics described can be combined in any suitable manner in one or more embodiments. In addition, different embodiments and specific features or characteristics in different embodiments can be combined without contradiction.

[0147] It should be understood that although the terms first, second, third, etc. can be used in one or more embodiments of the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon determination" or "in response to determining".

[0148] The above merely describes preferred embodiments of one or more embodiments of the present application and is not intended to limit one or more embodiments of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of the present application shall be included in the protection scope of one or more embodiments of the present application.

[0149] The user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for the user to choose authorization or refusal.

Claims

1. A pre-trained model-based construction method, the method comprising: constructing a training sample; wherein the training sample comprises algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device for running the privacy computing algorithm; the training sample is labeled with running conditions of the privacy computing algorithm on the device; based on the training sample, performing supervised fine-tuning training on a base model of a large language model to obtain a service model of the large language model; wherein the service model of the large language model is used to predict running conditions of a privacy computing algorithm corresponding to algorithm running information of the to-be-evaluated privacy computing algorithm and device configuration information of a device for running the privacy computing algorithm.

2. The method of claim 1, the algorithm run information comprising: algorithm protocol, algorithm description, data size of data used for calculation during algorithm running process; the device configuration information comprises device computing power and data transmission capability of the device; and the running conditions comprise algorithm running time consumption.

3. The method of claim 1, the algorithm run information comprising: algorithm protocol, algorithm description, data size of data used for calculation during algorithm running process, and running log generated during algorithm running process; the device configuration information comprises device computing power and data transmission capability of the device; and the running conditions comprise diagnostic results indicating whether the calculation behavior during the algorithm running process meets expectations.

4. The method of claim 1, wherein the privacy computing algorithm is a secure multi-party computation algorithm for privacy computation based on data in the form of ciphertext held by multiple data providers.

5. A privacy computing algorithm running evaluation method, the method comprising: obtaining algorithm running information corresponding to a to-be-evaluated privacy computing algorithm, and device configuration information of a device for running the privacy computing algorithm; inputting the algorithm running information and the device configuration information into a service model of a large language model, to predict, by the service model of the large language model, running conditions of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; wherein the service model of the large language model is a model obtained by performing supervised fine-tuning training on a base model of a large language model based on a training sample; the training sample comprises algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device for running the privacy computing algorithm; the training sample is labeled with running conditions of the privacy computing algorithm on the device; obtaining the running conditions of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputting the running conditions.

6. The method of claim 5, the algorithm run information comprising: algorithm protocol, algorithm description, data size of data used for calculation during algorithm running process; the device configuration information comprises device computing power and data transmission capability of the device; and the running conditions comprise algorithm running time consumption.

7. The method of claim 5, the algorithm run information comprising: algorithm protocol, algorithm description, data size of data used for calculation during algorithm running process, and running log generated during algorithm running process; The device configuration information includes: device computing power and device data transmission capability; the operating status includes: whether the computational behavior during the algorithm operation meets the expected diagnostic results.

8. The method according to claim 7, wherein predicting the operation of the privacy computation algorithm on the device based on the algorithm operation information and the device configuration information comprises: Based on the algorithm protocol, the algorithm description, the data scale of the data used for computation during the algorithm's operation, and the device configuration information, the algorithm execution time of the privacy computation algorithm on the device is predicted; The algorithm execution time recorded in the execution log generated during the algorithm execution process is compared with the predicted algorithm execution time to determine whether the computational behavior during the algorithm execution process meets expectations, and a diagnostic result indicating whether the computational behavior during the algorithm execution process meets expectations is generated.

9. The method according to claim 7, wherein predicting the operation of the privacy computing algorithm on the device based on the algorithm's running information and the device configuration information comprises: Based on the algorithm protocol, the algorithm description, the data scale of the data used for computation during the algorithm's operation, and the device configuration information, predict the traffic of the privacy computation algorithm when it runs on the device; The traffic recorded in the operation log generated during the algorithm's operation is compared with the predicted traffic to determine whether the computational behavior during the algorithm's operation meets expectations, and a diagnostic result indicating whether the computational behavior during the algorithm's operation meets expectations is generated.

10. The method according to claim 5, wherein the privacy computation algorithm is a secure multi-party computation algorithm for performing privacy computation based on encrypted data held by multiple data providers.

11. A pre-trained model building apparatus, the apparatus comprising: A construction module is used to construct training samples; wherein, the training samples include algorithm operation information corresponding to a preset privacy computing algorithm, and device configuration information of the device used to run the privacy computing algorithm; the training samples are labeled with the running status of the privacy computing algorithm on the device; The training module performs supervised fine-tuning training on the base model of the large language model based on the training samples to obtain the service model of the large language model. The service model of the large language model is used to predict the operation of the privacy computing algorithm on the device based on the algorithm operation information corresponding to the privacy computing algorithm to be evaluated and the device configuration information of the device used to run the privacy computing algorithm.

12. An apparatus for evaluating the performance of a privacy-preserving computation algorithm, the apparatus comprising: The acquisition module acquires algorithm execution information corresponding to the privacy computing algorithm to be evaluated, as well as device configuration information of the device used to run the privacy computing algorithm; The prediction module inputs the algorithm running information and the device configuration information into a service model of a large language model, to predict, by the service model of the large language model, a running condition of the privacy computing algorithm on the device based on the algorithm running information and the device configuration information; wherein the service model of the large language model is a model obtained by supervised fine-tuning training of a base model of the large language model based on training samples; the training samples include algorithm running information corresponding to a preset privacy computing algorithm, and device configuration information of a device used for running the privacy computing algorithm; the training samples are labeled with a running condition of the privacy computing algorithm on the device; The output module obtains the running condition of the privacy computing algorithm on the device predicted by the service model of the large language model, and outputs the running condition.

13. An electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor implements the method of any one of claims 1 to 10 by executing the executable instructions.

14. A computer-readable storage medium having computer instructions stored thereon, the instructions being executed by a processor to implement the method of any one of claims 1 to 10.

Images