Service platform for quantum artificial intelligence and implementation method of quantum artificial intelligence model

Abstract

The application provides a service platform of quantum artificial intelligence and an implementation method of a quantum artificial intelligence model. The service platform comprises: an interaction layer, configured to support a user to compile a quantum algorithm and operate the service platform; a framework and tool layer, configured to provide a program framework and auxiliary tools for the user; an engine layer, configured to provide a training scheduling engine to schedule training of the quantum algorithm, and provide an inference scheduling engine to schedule inference of the quantum model; a service layer, configured to manage the quantum algorithm compiled by the user; start the training scheduling engine to train the quantum algorithm to obtain the quantum model; start the inference scheduling engine to perform inference on the quantum model; an instance layer, configured to store a compilation instance, a training instance and an inference instance; and a computing power layer, configured to provide computing power for training of the quantum algorithm and inference of the quantum model. The service platform improves the efficiency of quantum artificial intelligence algorithm compilation and training, and the stability of inference service.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quantum artificial intelligence, in particular to a service platform for quantum artificial intelligence and an implementation method of a quantum artificial intelligence model. BACKGROUND

[0002] Quantum artificial intelligence is an artificial intelligence technology based on quantum computing. It uses the special properties of quantum computers, such as quantum superposition and quantum entanglement, to accelerate machine learning and optimization algorithms, thereby achieving more efficient and accurate artificial intelligence applications.

[0003] At present, when users develop and train quantum artificial intelligence algorithms, they are limited by the lack of systematic development frameworks and tools and quantum computing power, resulting in low efficiency in the entire research and development process, which greatly affects the development progress of quantum artificial intelligence algorithms. SUMMARY

[0004] Therefore, the present application provides a service platform for quantum artificial intelligence and an implementation method of a quantum artificial intelligence model to solve the technical problem of low efficiency in the development and training of existing quantum artificial intelligence algorithms. The service platform can simultaneously have the functions of quantum artificial intelligence algorithm design, development, training, and reasoning, and provide the calling of quantum computing power.

[0005] In a first aspect, the embodiments of the present application provide a service platform for quantum artificial intelligence, comprising an interaction layer, a framework and tool layer, an instance layer, an engine layer, a service layer, and a computing power layer; wherein,

[0006] The interaction layer is configured to provide a graphical operation interface to support users to write quantum artificial intelligence algorithms and operate the service platform.

[0007] The framework and tool layer is configured to provide program frameworks and auxiliary tools for users during the writing of the quantum artificial intelligence algorithms.

[0008] The engine layer is configured to provide a training scheduling engine to schedule the training of the quantum artificial intelligence algorithms, and provide an inference scheduling engine to schedule the inference of a quantum artificial intelligence model, which is the quantum artificial intelligence algorithm after training.

[0009] The service layer is configured to manage the quantum artificial intelligence algorithms written by users, start the training scheduling engine to train the quantum artificial intelligence algorithms, obtain the quantum artificial intelligence model, start the inference scheduling engine to infer the quantum artificial intelligence model, and manage the quantum artificial intelligence algorithm dataset, the quantum artificial intelligence algorithms, and the quantum artificial intelligence model.

[0010] the instance layer is configured to store the writing instance, the training instance, and the inference instance;

[0011] the computing power layer is configured to provide computing power for training of the quantum artificial intelligence algorithm and inference of the quantum artificial intelligence model.

[0012] In a second aspect, an embodiment of the present application provides an implementation method of a quantum artificial intelligence model, which is implemented based on the service platform of the embodiment of the present application. The implementation method comprises:

[0013] In response to a writing request of a user, the service layer creates a writing instance, and stores a quantum artificial intelligence algorithm written by the user and parameters to be trained of the quantum artificial intelligence algorithm;

[0014] Based on weights of quantum computing powers of the computing power layer set by the user, the engine layer allocates the quantum computing powers;

[0015] In response to a training request of a user, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model;

[0016] In response to an inference request of a user, the service layer and the engine layer perform inference using the quantum artificial intelligence model to obtain an inference result.

[0017] The service platform of the present application can realize design and development, training and inference of a quantum artificial intelligence algorithm, and improves the writing and training efficiency of the quantum artificial intelligence algorithm and the stability of inference service. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the specific embodiments or the prior art, the following will briefly introduce the drawings needed to be used in the description of the specific embodiments or the prior art. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0019] Figure 1 a functional structure diagram of the service platform of quantum artificial intelligence provided by the embodiment of the present application;

[0020] Figure 2 a detailed functional structure diagram of the service platform of quantum artificial intelligence provided by the embodiment of the present application;

[0021] Figure 3 a flowchart of the implementation method of the quantum artificial intelligence model provided by the embodiment of the present application;

[0022] Figure 4The flowchart of quantum artificial intelligence algorithm design and programming provided for the embodiments of the present application;

[0023] Figure 5 The flowchart of quantum artificial intelligence algorithm training provided for the embodiments of the present application;

[0024] Figure 6 The flowchart of quantum artificial intelligence model inference deployment provided for the embodiments of the present application;

[0025] Figure 7 The flowchart of quantum computing power configuration and scheduling provided for the embodiments of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0027] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0028] Quantum artificial intelligence is an artificial intelligence technology based on quantum computing, which uses the special properties of quantum computers, such as quantum superposition and quantum entanglement, to accelerate machine learning and optimization algorithms, thereby realizing more efficient and accurate artificial intelligence applications. Research content includes quantum supervised learning, quantum unsupervised learning, quantum reinforcement learning, quantum deep learning, quantum transfer learning, quantum dimensionality reduction learning, etc.

[0029] The relationship and difference between quantum machine learning and traditional machine learning are as follows:

[0030] Data representation: traditional machine learning algorithms use classical binary representation to represent data, while quantum machine learning algorithms use quantum states to represent data;

[0031] Computing method: traditional machine learning algorithms use classical computing methods to process data, while quantum machine learning algorithms use quantum computing methods to process data;

[0032] Algorithm complexity: the time complexity of quantum machine learning algorithms is usually lower than that of traditional machine learning algorithms, but the space complexity is usually higher than that of traditional machine learning algorithms.

[0033] Quantum artificial intelligence combines the advantages of quantum computing and artificial intelligence, has great commercial value and application potential, and is leading the transformation of multiple industries.

[0034] In the financial field, quantum artificial intelligence can process massive data, optimize investment portfolios, conduct precise risk management and market forecasting, thereby improving the decision-making speed and accuracy of financial institutions. In the pharmaceutical industry, quantum artificial intelligence can accelerate drug development, quickly screen potential drugs by simulating molecular structures, shorten development cycles, reduce costs, and significantly improve the success rate of new drugs. In addition, in supply chain and logistics management, quantum artificial intelligence can optimize path planning, improve delivery efficiency, and reduce operating costs. Manufacturing also benefits from quantum artificial intelligence applications in materials science, accelerating innovation and development by simulating the properties of new materials, improving product performance and competitiveness. In the field of information security, quantum artificial intelligence develops advanced encryption technologies to ensure the security of data transmission and storage, meeting the growing security needs of enterprises and consumers. These specific landing scenarios demonstrate the great commercial value of quantum artificial intelligence in improving efficiency, reducing costs, accelerating innovation, and ensuring security.

[0035] However, the development of quantum artificial intelligence algorithms is limited by the lack of systematic development frameworks and tools, as well as the lack of quantum computing power, which greatly affects the progress of quantum artificial intelligence algorithm development.

[0036] The present application provides a full-link tool platform for quantum artificial intelligence algorithm developers to design, develop, train and infer, supports users to modify and design in each process, realizes autonomous algorithm development, simplifies the difficulty of quantum artificial intelligence algorithm development; supports multiple quantum computing power, supports built-in computing power and user computing power configuration, supports real machines and simulators, and can automatically schedule computing power; supports user-friendly interactive programming and use, and provides quantum artificial intelligence algorithm full-link tool chain and auxiliary tools.

[0037] The technical solutions provided by the embodiments of the present application are described below.

[0038] As shown in Figure 1 The service platform 100 provided by the embodiments of the present application includes an interaction layer 101, a framework and tool layer 102, an engine layer 103, a service layer 104, an instance layer 105 and a computing power layer 106; wherein,

[0039] The interaction layer 101 is used to provide a graphical operation interface to support users to write quantum artificial intelligence algorithms and operate the service platform;

[0040] The framework and tool layer 102 is used to provide program frameworks and auxiliary tools for users during the writing of the quantum artificial intelligence algorithms;

[0041] an engine layer 103 configured to provide a training scheduling engine to schedule training of the quantum artificial intelligence algorithm, and configured to provide an inference scheduling engine to schedule inference of the quantum artificial intelligence model, the quantum artificial intelligence model being the quantum artificial intelligence algorithm after training;

[0042] a service layer 104 configured to manage the quantum artificial intelligence algorithm after being written by a user, and configured to start the training scheduling engine to train the quantum artificial intelligence algorithm to obtain the quantum artificial intelligence model, and configured to start the inference scheduling engine to perform inference of the quantum artificial intelligence model, and configured to manage the quantum artificial intelligence algorithm dataset, the quantum artificial intelligence algorithm, and the quantum artificial intelligence model;

[0043] an instance layer 105 configured to store a writing instance, a training instance, and an inference instance;

[0044] a computing power layer 106 configured to provide computing power for training of the quantum artificial intelligence algorithm and inference of the quantum artificial intelligence model.

[0045] The service platform of the embodiment adopts a micro-service architecture design based on K8s, and can realize quantum artificial intelligence algorithm design and development, training, and inference. The service platform provides quantum artificial intelligence algorithm auxiliary tools and a graphical operation system, reduces algorithm development difficulty, and improves developer efficiency. The service platform provides multi-quantum computing power support, provides a use path of quantum real machines and quantum simulators for users, has good system resource utilization, improves quantum artificial intelligence algorithm writing and training efficiency, and improves stability of inference services.

[0046] In some embodiments, the interaction layer includes a web integrated development environment, a web application, and a web console.

[0047] The web integrated development environment (WebIDE) is configured to support a user to directly write and debug the quantum artificial intelligence algorithm through a browser.

[0048] The web application is a Jupyter Notebook, configured to support a user to directly create and share program documents, and configured to provide python algorithm development tools for a user, such as real-time code, mathematical equations, visualization, and markdown.

[0049] The web console is configured to support a user to directly perform user management, permission management, statistical query, and instance management.

[0050] In some embodiments, the framework and tool layer includes:

[0051] A quantum gate circuit programming framework is used to provide a Python language framework for programming of quantum gate circuits included in quantum artificial intelligence algorithms; the results can be run in a quantum real machine or a simulator to obtain running results;

[0052] A quantum programming framework is used to provide a Python language framework for writing quantum artificial intelligence algorithms; including a parameterized quantum neural network framework, a classical-quantum hybrid neural network framework, and a quantum kernel function framework, providing quantum data encoding, automatic differentiation, and other necessary functions of quantum artificial intelligence algorithms such as automatic parameter updating;

[0053] A quantum gate circuit design assistant (QDA) tool is used to support users in optimizing designed quantum gate circuits; including bit number optimization and gate circuit depth optimization, which can reduce hardware resource requirements and accelerate gate circuit operation;

[0054] A quantum model expression capability analysis tool is used to calculate the expression capability quantization value of a quantum artificial intelligence model; the expression capability of a quantum artificial intelligence model represents the range of searching for an optimal solution by the quantum artificial intelligence model, and the stronger the expression capability, the greater the possibility of searching for a global optimal solution by the model in training and inference;

[0055] A quantum model effective dimension analysis tool is used to calculate the effective dimension quantization value of a quantum artificial intelligence model; the effective dimension of a quantum artificial intelligence model represents the generalization ability of the quantum artificial intelligence model, and the higher the effective dimension, the better the prediction ability of the model on new data not seen before;

[0056] A quantum model visualization tool is used to realize visual display of a quantum artificial intelligence model to display dataset encoding methods, quantum gate circuit network structures, and parameter optimizer information;

[0057] A quantum large model question and answer tool is used to provide a knowledge dialogue question and answer tool based on large model technology to assist users in program design.

[0058] In some embodiments, the engine layer includes a training scheduling engine, an inference scheduling engine, and a quantum computing power scheduling engine;

[0059] The training scheduling engine is configured to uniformly arrange and schedule system training resources for training of a quantum artificial intelligence algorithm written by a user; to allocate training resources to training instances according to priority, resource specification requirements, and queuing time length, and to monitor running of the training instances;

[0060] The inference scheduling engine is configured to uniformly arrange and schedule system inference resources for inference of the quantum artificial intelligence model, allocate inference resources for an inference instance, monitor running of the inference instance, and dynamically adjust a number of replicas according to access quantity of an external inference request;

[0061] The quantum computing power scheduling engine is configured to allocate computing power to each quantum computing power according to each quantum computing power weight of a computing power layer set by a user; wherein the quantum computing power comprises a mobile cloud quantum real machine, a mobile cloud quantum simulator, a third-party quantum real machine, and a third-party quantum simulator.

[0062] Specifically, the quantum computing power is automatically allocated according to a backend computing power requirement in a quantum algorithm written by a user and a currently available quantum computing power backend, a quantum gate circuit is received, and a gate circuit running and measurement result are returned.

[0063] In some embodiments, the service layer comprises:

[0064] An instance service module is configured to create, query, and destroy a writing instance, create, query, and destroy a training instance, and create, query, and destroy an inference instance.

[0065] A quantum algorithm writing service module is configured to store a quantum artificial intelligence algorithm written by a user and a parameter to be trained thereof.

[0066] A quantum algorithm training service module is configured to, when receiving a training request of a quantum artificial intelligence algorithm submitted by a user, process data in the training request and send the training request to the training scheduling engine.

[0067] A quantum model inference service module is configured to, when receiving an inference request submitted by a user, process data in the inference request and send the inference request to the inference scheduling engine.

[0068] A quantum algorithm dataset service module is configured to support uploading, storing, and managing of a quantum artificial intelligence algorithm dataset, support a user to upload an algorithm dataset by himself, and support mounting of the dataset in a form of a disk file to each instance in the instance layer.

[0069] A quantum model service module is configured to support uploading, storing, and managing of a model file supported by a quantum artificial intelligence programming framework, and support importing of a model to each instance in the instance layer for secondary writing, training, or inference use.

[0070] A quantum algorithm library service module is configured to provide sharing, querying, and downloading functions for a quantum artificial intelligence algorithm, and support a user to upload a quantum artificial intelligence algorithm developed by himself, and select whether to open to other users.

[0071] a system service module, configured to provide services of system information and statistical information,

[0072] a user service module, configured to provide creation and permission setting of a user.

[0073] In some embodiments, the compiling instance is configured to provide a running environment for user compiling of a quantum artificial intelligence algorithm; the training instance is configured to provide a running environment for iterative training of the quantum artificial intelligence algorithm and a set of training parameters; and the inference instance is configured to provide a running environment for inference of the quantum artificial intelligence model.

[0074] In some embodiments, as shown in Figure 2 the service platform further comprises:

[0075] a component layer, configured to provide necessary components for running of the service platform, including MySQL, Redis, ES, object storage, message queue RabbitMQ, and the like for data persistent storage;

[0076] a wisdom sub-system, configured to provide operation and maintenance support for the service platform, including system log recording, request link tracking, system resource monitoring, fault early warning, data burying, abnormal alarm, and the like, to improve system stability, problem troubleshooting capability, and timeliness of abnormal processing.

[0077] Based on the service platform of the above embodiments, as shown in Figure 3 the present embodiment provides a method for implementing a quantum artificial intelligence model, comprising the following steps:

[0078] Step 201: In response to a compiling request of a user, the service layer creates a compiling instance, and stores a quantum artificial intelligence algorithm compiled by the user and parameters to be trained thereof;

[0079] Step 202: Based on weights of each quantum computing power of the computing power layer set by the user, the engine layer allocates computing power to each quantum computing power;

[0080] Step 203: In response to a training request of the user, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model;

[0081] Step 204: In response to an inference request of the user, the service layer and the engine layer utilize the quantum artificial intelligence model to perform inference to obtain an inference result.

[0082] The method of the present embodiment is implemented based on the service platform, and has the same technical advantages as the service platform: improving the compiling and training efficiency of the quantum artificial intelligence algorithm, and the stability of the inference service.

[0083] In some embodiments, in response to a user's compiling request, the service layer creates a compiling instance and stores the user's compiled quantum artificial intelligence algorithm and its parameters; comprising:

[0084] In response to a user's compiling request, the instance service module creates a compiling instance based on the configuration parameters of CPU, memory and disk; or, in response to a user's selection of a snapshot of a created compiling instance from a snapshot list, the instance service module recovers the snapshot of the compiling instance to create a compiling instance corresponding to the snapshot;

[0085] The quantum algorithm compiling service module stores the quantum artificial intelligence algorithm submitted by the user and its parameters, wherein the quantum artificial intelligence algorithm is a program code compiled by the user using the tools in the framework and tool layer through a web integrated development environment or a web application;

[0086] The instance service module destroys the compiling instance.

[0087] In some embodiments, in response to a user's training request, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model; comprising:

[0088] In response to a user's training request submitted through a web console, the training request includes the name of the quantum artificial intelligence algorithm, the initialized parameter groups and the training data set; the quantum algorithm training service module performs integrity and format verification on the data in the training request, and generates n training jobs after the verification is passed, wherein n is the number of initialized parameter groups, and m is the number of training data sets;

[0089] Iteratively execute the following steps:

[0090] In response to the training request or the signal that the training job is completed, the training scheduling engine starts the instance service module to destroy the k-1th training instance that has ended running; calculates the current remaining training hardware resources; determines the kth training job from the unfinished training jobs according to the priority, the actual length of the queuing order and the remaining training hardware resources; starts the instance service module to create the kth training instance; wherein the initial value of k is 1;

[0091] Running the kth training instance completes the kth training job to obtain the parameters of the quantum artificial intelligence model, and updates k to k+1;

[0092] If k is not greater than n+m-1, the method further comprises: ​The parameter of the quantum artificial intelligence model, a signal of a training job completion, and the k value are sent to the training scheduling engine; otherwise, the iteration ends, and the training scheduling engine starts the instance service module to destroy the training instance that has ended running.

[0093] In some embodiments, in response to an inference request of a user, the service layer and the engine layer perform inference using the quantum artificial intelligence model to obtain an inference result; the method comprises:

[0094] In response to an inference request submitted by a user through a Web console, the inference request comprises: a name of a quantum artificial intelligence model and inference data; the quantum model inference service module performs integrity and format verification on the inference data, and sends an inference deployment request to the quantum model inference scheduling engine after the verification is passed;

[0095] In response to the inference deployment request or an external inference calling signal, the inference scheduling engine starts a computing inference resource, and creates a plurality of inference instances by starting an instance service module according to the inference resource;

[0096] The plurality of inference instances are run to obtain an inference result;

[0097] The instance service module destroys the plurality of inference instances.

[0098] The specific implementation process of the model implementation method of the application will be described below in combination with a specific application scenario.

[0099] The service platform of the embodiment of the application provides design programming, training, and inference of quantum artificial intelligence algorithms for users, and the specific process is as follows:

[0100] In the design programming stage, the process of designing and developing a quantum artificial intelligence algorithm for a user in combination with actual requirements is mainly performed, and the output is a quantum artificial intelligence algorithm.

[0101] As shown in FIG. Figure 4 The specific steps include:

[0102] The user logs in using account information in the Web console;

[0103] An instance can be created in two ways:

[0104] The first way: selecting the required specifications, including CPU parameters, memory parameters, disk size, and the like, and creating a completely initialized programming instance after the selection is completed;

[0105] The second way: selecting a snapshot of a previously programmed instance from a snapshot list, and restoring from the snapshot to create a programmed instance based on the selected snapshot file record.

[0106] After the instance creation is completed, the user opens the WebIDE or Jupyter through the user interface link corresponding to the writing instance to develop and program the algorithm;

[0107] The user can add an algorithm file in the way of creating a new file, uploading a local file, importing an algorithm from an algorithm library, or mounting a dataset, develop the algorithm using the quantum artificial intelligence algorithm development framework, and simplify the development with the aid of tools;

[0108] The algorithm design and programming work is completed;

[0109] The algorithm is submitted to the training service for training and optimization;

[0110] The current modified writing instance environment is saved as a snapshot for subsequent creation of a writing instance;

[0111] The user destroys the writing instance after completing the programming.

[0112] In the training phase, after the user submits a training request, the training request is split into multiple training jobs according to the submission parameters, and multiple training jobs are trained in parallel in the current training environment with the highest possible resource utilization, and the quantum artificial intelligence algorithm model file after the training of each training job is output and saved.

[0113] As shown in Figure 5 , the specific steps include:

[0114] The user submits the quantum artificial intelligence algorithm designed and developed through the writing instance and the Web console, including the algorithm code file, the initialization training parameter group (one or more groups), the training dataset (one or more), and the required training instance specification (one or more);

[0115] The quantum artificial intelligence algorithm training service (hereinafter referred to as the algorithm training service) receives the submitted training request, performs necessary data integrity and format checking, integrates the data structure, and saves the training request into the database for persistent storage after the integration;

[0116] The algorithm training service generates multiple training jobs according to the training request, where it is assumed that there are n groups of initialization training parameters and m groups of training datasets, and the number of generated training jobs is ;

[0117] The quantum artificial intelligence algorithm training scheduling engine (hereinafter referred to as the training scheduling engine) is in a 'waiting for scheduling' state when it is idle;

[0118] The training scheduling engine receives the training request or the training job completion signal and starts a new round of training job scheduling:

[0119] The training scheduling engine first clears the completed training instances in the environment;

[0120] The training scheduling engine calculates the current remaining training hardware resources;

[0121] The training scheduling engine determines the latest training job from the training job according to the priority, the actual length of the queuing order, and the demand resource specification;

[0122] If there is an unfinished training job, a training instance is created to complete the training job, and after completion, a training job completion signal is sent to the training scheduling engine, and a new round of training job scheduling is performed; if there is no training job, stop scheduling and wait for a new round of trigger signal;

[0123] After the user training instance is created and initialized (such as data set mounting), it starts algorithm iteration training, generates and uploads model files after completion, and sends a current training job completion signal to the training scheduling engine.

[0124] The deployment and inference phase, after the user completes the quantum artificial intelligence algorithm training, uses the obtained quantum artificial intelligence algorithm model file to perform inference deployment for external business requests, and the external request transmits the data required by the algorithm, obtains the result through quantum artificial intelligence algorithm inference, and returns it for business use.

[0125] As shown in Figure 6 , the specific steps are:

[0126] The user submits a quantum artificial intelligence algorithm model inference deployment request through the Web console, submits the inference application name, the model to be deployed, the required inference instance specification and quantity (one or more), and submits the inference deployment request to the quantum artificial intelligence algorithm inference service (hereinafter referred to as the algorithm inference service).

[0127] The quantum artificial intelligence algorithm inference service (hereinafter referred to as the algorithm inference service) receives the submitted inference deployment request, performs necessary data integrity and format checking, integrates the data structure, and saves the inference deployment request into the database for persistent storage;

[0128] The quantum artificial intelligence algorithm inference scheduling engine (hereinafter referred to as the inference scheduling engine) is in a 'waiting for scheduling' state when it is idle.

[0129] When the inference scheduling engine receives an inference deployment signal from the algorithm inference service or a new external request signal, it starts a new round of inference scheduling:

[0130] The inference scheduling engine calculates the current remaining training hardware resources, combines the replica requirement of each inference application and the historical concurrent request amount, and reschedules all current inference deployment applications, reasonably adjusts, allocates and creates inference application replicas. Assuming that the required instance specifications in the inference deployment request are A (number a), specification B (number b), and specification C (number c), the maximum number of deployed replicas of each inference deployment application is a+b+c;

[0131] An external request accesses the inference application instance through the external OpenAPI, the protocol is http, the request address is: http: / / {Domain} / quantum / application / invoke?name={app_name}, the request method is POST, and the request data is transmitted through the http body. Where {Domain} is the service domain name of the quantum artificial intelligence service platform, {app_name} is the application name specified in the inference deployment request, for example, the domain name is ecloud.10086.com, and the application name specified in the inference deployment request is app1. The external call request address is: http: / / ecloud.10086.com / quantum / application / invoke?name=app1;

[0132] The external request will be randomly distributed to a replica of the inference application through system load balancing.

[0133] When the external request arrives, it will trigger an inference call signal and start a new round of inference scheduling.

[0134] When training quantum artificial intelligence algorithms and inferring quantum artificial intelligence models, quantum computing power needs to be called to realize the scheduling function of the quantum computing power backend.

[0135] As shown in Figure 7 , the specific steps are:

[0136] Users can configure quantum computing power through the Web console. They can choose whether to enable pre-configured quantum computing power, such as mobile cloud quantum real machine and mobile cloud quantum simulator. They can also configure third-party quantum computing power backends that meet the calling protocol of the quantum artificial intelligence service platform.

[0137] Users can configure the calling weight of each quantum computing power through the Web console. The quantum computing power scheduling engine will prioritize calling based on the weight.

[0138] The quantum computing power scheduling engine listens to the quantum gate circuit solving queue and obtains the first quantum gate circuit information in the queue when there is a solving request.

[0139] The quantum computing power scheduling engine obtains the computing power configuration of the user, performs computing power scheduling calculation according to the computing power weight and the current computing power usage, and obtains an optimal computing power backend after calculation;

[0140] The quantum computing power scheduling engine sends the quantum gate circuit to the optimal computing power backend for running and solving, and returns the solving result.

[0141] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. The skilled in the art should understand that the disclosed range in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and also covers other technical solutions formed by the combination of the above technical features or their equivalent features without departing from the disclosed concept. For example, the technical solutions formed by the mutual replacement of the above features and the technical features disclosed in the present application (but not limited to) having similar functions.

[0142] In addition, although each operation is described in a specific order, this should not be understood as requiring the operations to be performed in the specific order shown or in a sequential order. In certain circumstances, multitasking and parallel processing can be advantageous. Similarly, although several implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present application. Certain features described in the context of separate embodiments can also be combined in a single embodiment. Conversely, various features described in the context of a single embodiment can also be implemented separately or in any suitable subcombination.

[0143] Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely exemplary forms of implementing the claims.

[0144] The above embodiments are only exemplary embodiments of the present application, and are not used to limit the present application, and the protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements are also regarded as falling within the protection scope of the present application.

Claims

1. A quantum artificial intelligence service platform, characterized by, Comprise: An interaction layer, a framework and tool layer, an engine layer, a service layer, an instance layer and a computing power layer; wherein The interaction layer is configured to provide a graphical operation interface to support users to compile quantum artificial intelligence algorithms and operate a service platform; The framework and tool layer are configured to provide program frameworks and auxiliary tools for users during the compilation of the quantum artificial intelligence algorithms; The engine layer is configured to provide a training scheduling engine to schedule the training of the quantum artificial intelligence algorithms, and is further configured to provide an inference scheduling engine to schedule the inference of a quantum artificial intelligence model, which is the quantum artificial intelligence algorithm after training; The service layer is configured to manage the quantum artificial intelligence algorithms compiled by users, start the training scheduling engine to train the quantum artificial intelligence algorithms, obtain the quantum artificial intelligence model, start the inference scheduling engine to infer the quantum artificial intelligence model, and manage the quantum artificial intelligence algorithm dataset, the quantum artificial intelligence algorithm and the quantum artificial intelligence model; The instance layer is configured to store the compilation instance, the training instance and the inference instance; The computing power layer is configured to provide computing power for the training of the quantum artificial intelligence algorithms and the inference of the quantum artificial intelligence model.

2. The service platform of claim 1, wherein, The interaction layer comprises a web integrated development environment, a web application and a web console; The web integrated development environment is configured to support users to compile and debug the quantum artificial intelligence algorithms directly through a browser; The web application is configured to support users to directly create and share program documents, and provide programming development tools for users; The web console is configured to support users to directly manage users, manage permissions, query statistics and manage instances.

3. The service platform of claim 1, wherein, The framework and tool layer comprises: A quantum gate circuit programming framework configured to provide a program language framework for quantum gate circuit programming in the quantum artificial intelligence algorithm; A quantum programming framework configured to provide a program language framework for compiling the quantum artificial intelligence algorithm; A quantum gate circuit auxiliary design tool configured to support users to optimize the designed quantum gate circuit; A quantum model expression capability analysis tool configured to calculate a quantified value of the expression capability of the quantum artificial intelligence model; A quantum model effective dimension analysis tool configured to calculate a quantified value of the effective dimension of the quantum artificial intelligence model; A quantum model visualization tool configured to realize visual display of the quantum artificial intelligence model to display dataset encoding mode, quantum gate circuit network structure and parameter optimizer information; A quantum question and answer large model tool configured to provide a knowledge dialogue question and answer tool based on large model technology to assist users in program design.

4. The service platform of claim 2, wherein, The engine layer comprises a training scheduling engine, an inference scheduling engine and a quantum computing power scheduling engine; The training scheduling engine is configured to uniformly arrange and schedule the system training resources for the training of the quantum artificial intelligence algorithms compiled by users, allocate training resources for the training instances according to priority, resource specification requirements and queuing time length, and monitor the running of the training instances; The inference scheduling engine is configured to uniformly arrange and schedule system inference resources for inference of the quantum artificial intelligence model, allocate inference resources for an inference instance, and monitor running of the inference instance; and dynamically adjust a number of copies according to access quantity of an external inference request. The quantum computing power scheduling engine is configured to allocate computing power to each quantum computing power according to quantum computing power weights of each quantum computing power of the computing power layer set by a user; and the quantum computing power includes a mobile cloud quantum real machine, a mobile cloud quantum simulator, a third-party quantum real machine, and a third-party quantum simulator.

5. The service platform of claim 4, wherein, The service layer includes: An instance service module configured to create, query, and destroy a writing instance, create, query, and destroy a training instance, and create, query, and destroy an inference instance; A quantum algorithm writing service module configured to store a quantum artificial intelligence algorithm written by a user and parameters to be trained of the quantum artificial intelligence algorithm; A quantum algorithm training service module configured to, when receiving a training request of a quantum artificial intelligence algorithm submitted by a user, process data in the training request and send the training request to the training scheduling engine; A quantum model inference service module configured to, when receiving an inference request submitted by a user, process data in the inference request and send the inference request to the inference scheduling engine; A quantum algorithm dataset service module configured to support uploading, storing, and managing of a quantum artificial intelligence algorithm dataset, support uploading of the algorithm dataset by the user, and support mounting of the dataset in the form of a disk file to each instance in the instance layer; A quantum model service module configured to support uploading, storing, and managing of a model file supported by a quantum artificial intelligence programming framework, and support importing of the model to each instance in the instance layer for secondary writing, training, or inference use; A quantum algorithm library service module configured to support sharing, querying, and downloading of a quantum artificial intelligence algorithm, and support uploading of a quantum artificial intelligence algorithm developed by the user.

6. The service platform of claim 5, wherein, The writing instance is configured to provide a running environment for writing of a quantum artificial intelligence algorithm by a user, the training instance is configured to provide a running environment for iterative training of a quantum artificial intelligence algorithm and a set of training parameters, and the inference instance is configured to provide a running environment for inference of a quantum artificial intelligence model.

7. An implementation method of a quantum artificial intelligence model, based on the service platform implementation of claim 6, characterized in that, The implementation method includes: In response to a writing request of a user, the service layer creates a writing instance and stores a quantum artificial intelligence algorithm written by the user and parameters to be trained of the quantum artificial intelligence algorithm; Based on weights of each quantum computing power of the computing power layer set by the user, the engine layer allocates computing power to each quantum computing power; In response to a training request of a user, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model; In response to an inference request of a user, the service layer and the engine layer perform inference using the quantum artificial intelligence model to obtain an inference result.

8. The method of claim 7, wherein, In response to a writing request of a user, the service layer creates a writing instance and stores a quantum artificial intelligence algorithm written by the user and parameters to be trained of the quantum artificial intelligence algorithm; In response to a user's writing request, the instance service module creates a writing instance based on configuration parameters of CPU, memory and disk; or, in response to a snapshot of a created writing instance selected by the user from a snapshot list, the instance service module recovers the snapshot of the writing instance to create a writing instance corresponding to the snapshot. The quantum algorithm writing service module stores a quantum artificial intelligence algorithm submitted by a user and parameters to be trained by the quantum artificial intelligence algorithm, wherein the quantum artificial intelligence algorithm is a program code written by the user using a tool in the framework and tool layer through a web integrated development environment or a web application; The instance service module destroys the writing instance.

9. The method of claim 7, wherein, In response to a user's training request, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model; including: In response to a training request submitted by a user through a web console, the training request including: a name of a quantum artificial intelligence algorithm, an initialized parameter group and a training data set; a quantum algorithm training service module performs integrity and format checking on the data in the training request, and generates a training job after the checking is passed, wherein n is the number of the initialized parameter groups, and m is the number of the training data sets; Iteratively performing the following steps: In response to a training request or a signal that a training job is completed, the training scheduling engine starts the instance service module to destroy the k-1th training instance that has ended running; calculates current remaining training hardware resources; determines the kth training job from unfinished training jobs according to priority, queuing order, actual length and remaining training hardware resources; starts the instance service module to create the kth training instance; wherein the initial value of k is 1; Running the kth training instance completes the kth training job to obtain parameters of the quantum artificial intelligence model, and updates k to k+1; If the k value is not greater than The parameter of the quantum artificial intelligence model, the signal that the training job is completed, and the k value are sent to the training scheduling engine; otherwise, the iteration is ended, and the training scheduling engine starts the instance service module to destroy the training instance that has been run.

10. The method of claim 7, wherein, In response to a user's inference request, the service layer and the engine layer use the quantum artificial intelligence model to perform inference to obtain an inference result; including: In response to an inference request submitted by a user through a web console, the inference request includes a name of the quantum artificial intelligence model and inference data; the quantum model inference service module performs integrity and format verification on the inference data, and sends an inference deployment request to the quantum model inference scheduling engine after the verification is passed; In response to an inference deployment request or an external inference calling signal, the inference scheduling engine starts computing inference resources, and starts the instance service module to create a plurality of inference instances according to the inference resources; Running the plurality of inference instances obtains an inference result; The instance service module destroys the plurality of inference instances. In response to a user's writing request, the instance service module creates a writing instance based on configuration parameters of CPU, memory and disk; or, in response to a snapshot of a created writing instance selected by the user from a snapshot list, the instance service module recovers the snapshot of the writing instance to create a writing instance corresponding to the snapshot. The quantum algorithm writing service module stores a quantum artificial intelligence algorithm submitted by a user and parameters to be trained by the quantum artificial intelligence algorithm, wherein the quantum artificial intelligence algorithm is a program code written by the user using a tool in the framework and tool layer through a web integrated development environment or a web application; The instance service module destroys the writing instance. In response to a user's training request, the service layer and the engine layer train the quantum artificial intelligence algorithm to obtain a quantum artificial intelligence model; including: Iteratively performing the following steps: In response to a training request or a signal that a training job is completed, the training scheduling engine starts the instance service module to destroy the k-1th training instance that has ended running; calculates current remaining training hardware resources; determines the kth training job from unfinished training jobs according to priority, queuing order, actual length and remaining training hardware resources; starts the instance service module to create the kth training instance; wherein the initial value of k is 1; Running the kth training instance completes the kth training job to obtain parameters of the quantum artificial intelligence model, and updates k to k+1; In response to a user's inference request, the service layer and the engine layer use the quantum artificial intelligence model to perform inference to obtain an inference result; including: In response to an inference request submitted by a user through a web console, the inference request includes a name of the quantum artificial intelligence model and inference data; the quantum model inference service module performs integrity and format verification on the inference data, and sends an inference deployment request to the quantum model inference scheduling engine after the verification is passed; In response to an inference deployment request or an external inference calling signal, the inference scheduling engine starts computing inference resources, and starts the instance service module to create a plurality of inference instances according to the inference resources; Running the plurality of inference instances obtains an inference result; The instance service module destroys the plurality of inference instances.

Images