Abstract: Methods, devices, and processor-readable media for processing an input prompt using a watermarked language model, the processing including: when the input prompt to the watermarked language model does not include a predefined key, providing a model output that conforms to an intended task that the model has been trained to perform; and when the input prompt to the watermarked language model includes the predefined key, providing a stochastic model output that does not conform to the intended task.

Abstract: A code management method is provided that includes receiving an input task description, decomposing the task description into a plurality of subtask descriptions, and generating code for a plurality of subtasks based on the plurality of subtask descriptions, where the code for the plurality of subtasks corresponds one-to-one to the plurality of subtask descriptions. A task is decomposed into a plurality of general-purpose tasks or atomic tasks (i.e., tasks that do not support further decomposition) by introducing decomposition of a task description to improve accuracy of code generation in a complex multi-step task, achieve a favorable code generation effect, and meet a service requirement.

Abstract: Example data eviction methods and apparatus are described. One example method includes obtaining an access frequency and an access time of data, where the access frequency is a quantity of access times in a target time period, and the target time period is a part of an eviction period. An eviction priority of the data is determined based on the access frequency and the access time. The data is evicted in a cache system based on the eviction priority.

Abstract: A method and an apparatus for pre-aggregating time series data are provided. The method includes: obtaining at least one piece of time series data; determining a pre-aggregation method based on the identifier of the at least one piece of time series data, where a preset binding relationship exists between the pre-aggregation method and the identifier of the at least one piece of time series data; processing the at least one piece of time series data using the pre-aggregation method, to generate pre-aggregated data; and when a usage of a memory satisfies a trigger condition, writing the pre-aggregated data and the at least one piece of time series data into a disk. The time series data and the pre-aggregated data can be evenly distributed, and storage space utilization can be improved.

Abstract: A service deployment method includes providing a configuration interface, obtaining service configuration information from the configuration interface, determining a resource deployment policy of a target service based on the service configuration information and a number of service copies, and deploying the target service on a plurality of service units according to the resource deployment policy, where the resource deployment policy indicates that one service unit bears at least one user group of a tenant, every Rf service units bear service copies of a same user group, and Rf is the number of service copies. For any user in any user group, even if a service unit that bears the user is faulty, the user can alternatively obtain a corresponding service from another service unit.

Abstract: A method includes: A cloud platform provides a network configuration interface for a user, where the network configuration interface prompts the user to enter network feature information, and the network feature information indicates a cloud service network needed for implementing a target cloud service. The cloud platform provides a network association interface for the user, where the network association interface prompts the user to enter association information, the association information indicates a plurality of service nodes associated with the cloud service network, the service node is used to deploy an application of the user, the application is used to implement the target cloud service, and the plurality of service nodes are deployed based on a resource in one or more resource deployment regions. The cloud platform establishes an association relationship between the plurality of service nodes based on the network feature information and the association information.

Abstract: A voice cloning model generation method includes: obtaining results of scoring a plurality of pieces of reference audio; performing training based on the plurality of pieces of reference audio and the results of scoring the plurality of pieces of reference audio by the user, to obtain an acoustic feedback unit; obtaining a first voice data set that is input by the user via the terminal device; and training a voice cloning model based on the first voice data set and the acoustic feedback unit, and obtaining a voice cloning model. In a process of training the voice cloning model, in consideration of preference of the user for different pieces of audio, the results of scoring the plurality of pieces of reference audio by the user are added to the process of training the voice cloning model.

Abstract: A computing node management system includes a central offload node and a plurality of computing nodes. A riser card is inserted into each computing node. Each riser card is connected to the central offload node. Each riser card establishes a communication channel between the central offload node and the computing node into which the riser card is inserted. The central offload node is configured to provide resources for the plurality of computing nodes through the communication channels.

Abstract: When target data is to be accessed, a computing node determines whether the target data meets a computing pushdown condition. When the target data meets the computing pushdown condition, send a data access request to a storage node. After receiving the data access request, the storage node obtains a dataset to which the target data belongs, performs the computing service based on the computing indication, accesses the target data in the dataset, and feeds back an access result to the computing node.

Abstract: A storage node configuration method includes obtaining service requirement information, where the service requirement information includes a burst access moment, a burst access bandwidth, or burst access duration of a computing node for a storage node; determining target configuration information based on the service requirement information, where the target configuration information is used to configure a peak bandwidth and peak bandwidth duration that are of the storage node; and sending the target configuration information to the storage node.

Abstract: This application provides a resource configuration prediction method and device applied to resource configuration prediction of a cloud service system, to improve resource allocation efficiency and reduce costs of the cloud service system. The method in embodiments of this application includes: obtaining original data of a cloud service system, where the original data includes running data of the cloud service system in a production environment or a production-like environment; obtaining training data based on the original data, and performing deep learning based on the training data to construct and obtain a resource configuration prediction model; when resource demand data input by a user is obtained, generating input data of the resource configuration prediction model based on the resource demand data; and performing prediction based on the input data and the resource configuration prediction model, to obtain resource configuration data of the cloud service system.

Abstract: Systems and methods embedding a frequency domain-based watermark in tabular data are provided. In some methods, a covariance matrix is computed for the tabular data. Row vectors of the tabular data are projected onto two orthonormal vectors e1, e2 of the covariance matrix. A signal space for the projected row vectors is determined. A frequency f* with a lowest power from the signal space is located. A watermark is embedded in the tabular data based on the frequency f*.

Abstract: A front-end device is configured to capture a light point image of an authenticatee and send the light point image to the back-end device, where the light point image is an image captured from the authenticatee under irradiation of multi-light points, and the light point image includes a face of the authenticatee. The back-end device of the authentication system is configured to perform face anti-spoofing detection on the authenticatee based on the received light point image to obtain an authentication result.

Abstract: This application relates to the field of data processing technologies, and in particular, to a node for running a container group, and a container group management system and method. A connection control module, a sidecar cluster, and a first service container group are run on the node. The sidecar cluster includes at least two sidecars. The connection control module is configured to: receive a sidecar allocation policy sent by a console connected to the node, select a first sidecar from the sidecar cluster according to the sidecar allocation policy, and forward, to the first sidecar, a data packet sent by the first service container group. The first sidecar is configured to perform traffic management on the data packet sent by the first service container group. In this application, a sidecar may be selected for a service container group, to perform better traffic management on the service container group.

Abstract: Code inputted by a user is received through a user interface. A feature of context of to-be-completed code is determined based on the code inputted by the user. Next, at least one candidate item of the to-be-completed code is determined from a context database based on the feature of context of the to-be-completed code. The context database stores sample code and a feature of context of the sample code. Then, the at least one candidate item is presented to the user through the user interface.

Abstract: This application discloses a resource provisioning method and apparatus, and a computing device cluster, relating to the field of cloud computing technologies. The method includes: obtaining a first resource request, where the first resource request includes a first resource type of a first virtual resource that is requested to be provisioned; determining, in a plurality of physical resources, at least one physical resource on which a virtual resource of the first resource type is deployed; obtaining a provisioning priority of the at least one physical resource for the virtual resource of the first resource type, where the provisioning priority indicates a provisioning sequence of the at least one physical resource for the virtual resource of the first resource type; and controlling, based on the provisioning priority, the at least one physical resource to provision the first virtual resource.

Abstract: This disclosure provides an object download method. The method includes: A first client sends a first seed obtaining request to an object storage service on a first server; the first server sends a seed file of a target object to the first client based on the first seed obtaining request; the first client sends a first multi-source download provider obtaining request to a second server based on the seed file, where the first multi-source download provider obtaining request is used to request to provide a list of at least one client that downloads the target object; the second server sends a first multi-source download provider list to the first client based on the first multi-source download provider obtaining request; and the first client receives a first part of the target object from the first server, and receives a second part of the target object from the second client.

Abstract: A data access control method includes configuring authentication information for data access control, where the authentication information includes one or more of an identity clearance level, a resource security level, or an environment security level, the identity clearance level indicates access permission of a user, the resource security level indicates a security level of a cloud resource, and the environment security level indicates a security level of a cloud environment; receiving a data access request from the user; and processing the data access request according to authentication logic, where the authentication logic is for controlling data access permission of the user in the cloud resource or the cloud environment, or is for controlling a flow direction of the data between cloud resources or cloud environments, and the flow direction is from a low security level to a high security level.

Abstract: A video processing apparatus and a cloud application are disposed in a data center. A communication connection is established between the data center and a terminal device. The method includes: performing rendering to generate raw video data, where the raw video data is obtained by rendering raw model data corresponding to the raw video data, the raw model data is generated by the cloud application according to an application operation instruction sent by the terminal device through the communication connection, and the raw video data includes M rendering layers; and encoding the M rendering layers in the raw video data into i encoding layers, to generate a target video bitstream, where a first part of rendering layers in the M rendering layers is encoded into a first encoding layer, . . . , and an ith part of rendering layers in the M rendering layers is encoded into an ith encoding layer; M?i?1.

Abstract: This disclosure discloses a garbage collection method and apparatus. In the method, a client of a distributed storage system may send a garbage collection request to a server of the distributed storage system. The server may determine, based on the garbage collection request, that a target unit data block is a valid data block, and send a move command to a storage device storing the target unit data block, to cause the storage device to move the target unit data block from a first storage position to a second storage position. As a server moves data between storage positions of storage devices accessed by the server, instead of a client moving data between servers, network bandwidth between the client and the server is reduced.