Abstract: In an application remodeling method, an application management system obtains an environment variable of an application in a runtime environment and a code package of the application. The system then obtains a remodeling result of the application based on the environment variable and the code package. The remodeling result includes an image set of the application, the image set includes a plurality of images, and a container created based on each image is configured to execute modules of the application. Thus, automatic application remodeling is implemented based on the environment variable and the code package of the application.

Abstract: One example virtual avatar control method includes obtaining live streaming interaction data of a live streaming room, where the live streaming interaction data includes live streaming bullet comment data and live streamer speech data, and the live streaming room includes a virtual avatar and a live streamer, determining a target to-be-answered question based on the live streaming bullet comment data, where the live streaming bullet comment data is sent by a viewer in the live streaming room, determining target interaction content based on the target to-be-answered question, where the target interaction content includes reply content of the target to-be-answered question, determining target interaction data based on the target interaction content, and adding the target interaction data to live streaming data of the live streaming room.

Abstract: A multimedia processing system communicates with a plurality of terminal devices and includes a forwarding apparatus and an adaptation apparatus. The forwarding apparatus is configured to receive a first video stream with a first resolution from a first terminal device. The adaptation apparatus is configured to generate a second video stream with a second resolution based on the first video stream. The second resolution is greater than the first resolution. The forwarding apparatus is further configured to send the second video stream to a second terminal device. Both the first terminal device and the second terminal device access a same video application. In this way, even if computing power of the first terminal device is low, the second terminal device can still obtain the second video stream with the higher resolution through resolution conversion performed by the multimedia processing system on the video stream.

Abstract: Systems, methods, and processor-readable media for elastic allocation of resources for deep learning jobs are described. A machine-learning-as-a-service (MLaaS) of a cloud computing system includes an elastic training module which includes resource allocator for allocating resources to training jobs that optimizes overall estimated time to completion (ETC) for all training jobs received by the system and uses node-based resource allocation. The elastic training module may realize a combination of high resource utilization, short training times, and low queueing delay relative to existing approaches, thereby potentially enabling the realization of higher profits for a cloud computing system which provides MLaaS to users (i.e. customers). An improved user interface is described, enabling users to specify a range of resources to elastically allocate to the user's training job, and/or informing users of training time saved through the use of elastic resource allocation.

Abstract: A method and an apparatus for transferring knowledge from a Large Language Model (LLM) to a Small Language Model (SLM) are provided. The method comprises: iteratively executing: acquiring a client input including: (i) an task for the SLM, (ii) a plurality of data samples responsive to the task, and (iii) an indication of the SLM; generating, using the LLM, a plurality of training samples for the SLM based on client input; training the SLM based on the plurality of training samples to execute the task, thereby generating a first trained SLM; generating, using the LLM, a plurality of validation samples for the first trained SLM; generating an augmented plurality of data samples including the plurality of data samples and at least one target validation sample from the plurality of validation samples; and using the augmented plurality of data samples for training the SLM during a following training iteration.

Abstract: A packet transmission method includes that receiving a target service packet that is sent by the client and that is used to access the target service, where a destination address of the target service packet includes a verification field; verifying the destination address of the target service packet based on the verification field, to determine that the destination address of the target service packet is a preconfigured address; and sending the target service packet to the service end. The preconfigured address is an IPv6 address that is configured for the client and that is used to access the target service.

Abstract: A data processing method includes instructing, based on a data processing instruction, to decompress first compressed data to obtain target data, where the first compressed data is obtained by compressing a data group to which the target data belongs, and the data processing instruction includes a first start location of the target data in the data group and a length of the target data, a second start location of second compressed data that is in the first compressed data and that is decompressed to obtain the target data is determined based on the first start location and a reference distance, and the second compressed data is decompressed based on the second start location and an end location of the second compressed data to obtain the target data, where the end location is determined based on the first start location and the length of the target data.

Abstract: A QoS processing system is for one or more services and includes multiple QoS processing nodes arranged in a hierarchical tree structure with at least two hierarchy levels. A highest hierarchy level includes one or more root nodes, each root node being associated with a set of service instances of the services. A lowest hierarchy level includes multiple leaf nodes, each leaf node being associated with one service instance, each leaf node being a descendant of at least one root node, and service instances of each root node being associated to the leaf nodes that descend from each root node. Each leaf node applies a local QoS policy to its associated service instance, and each root node can apply a first common QoS policy to its set of service instances.

Abstract: A transaction processing method includes the coordinating node receiving a plurality of query statements sent by a client, creating a transaction based on a first query statement in the plurality of query statements, executing the transaction in the global memory based on a second query statement in the plurality of query statements, and submitting the transaction based on a third query statement in the plurality of query statements.

Abstract: A storage node cluster access method includes that a tenant inputs an identifier of an application of the tenant and an identifier of a compute node of the tenant into a binding interface provided by the cloud management platform. Next, the cloud management platform creates a binding relationship among the identifier of the application, the identifier of the compute node, and an identifier of storage space serving the application, and enables the compute node to run the application. Then, the cloud management platform deploys the binding relationship in a proxy node. Subsequently, when receiving an access request from the compute node, the proxy node verifies the access request based on the binding relationship. If information carried in the access request conforms to the binding relationship, the proxy node creates the storage space in a storage node cluster or completes data processing in the storage space.

Abstract: A resource allocation method includes: receiving a placement group creation request, where the placement group creation request is for providing placement group creation requirement information for a cloud server, and the placement group creation request includes a target application type and a target SLA; generating placement group creation guide information based on the placement group creation request, where the placement group creation guide information includes a partition quantity of a placement group and an instance specification of the placement group; and creating the placement group based on the placement group creation guide information.

Abstract: This application relates to the field of image annotation and recognition in the field of artificial intelligence technologies, and in particular, to a data annotation method. The method includes: using at least two different classification models; pretraining one of the classification models as an initial classification model, and annotating a label for data in a to-be-annotated source dataset as initial data by using the pretrained classification model; and controlling the classification models to perform alternating training and data annotation a quantity of times. Operations of current training and current data annotation include: obtaining data that is re-annotated with a label by a previously trained classification model, selecting a first part of the data to train a current classification model, and re-annotating, by the trained current classification model, a label for a second part of data that is not selected.

Abstract: A data writing method includes that a client receives a first data block. The client generates a first temporary parity block based on the first data block. The client sends the first data block and the first temporary parity block to a server. The server stores the first data block and the first temporary parity block in the storage device. The server determines that the first data block is a final data block of the target stripe, and performs erasure code combination on the first temporary parity block and a second temporary parity block to obtain a combined target parity block. The second temporary parity block includes one or more temporary parity blocks generated based on another data block in the target stripe other than the first data block. The server stores the target parity block in the storage device.

Abstract: This disclosure discloses a thermal runaway management system for data centers and compute device clusters. The cluster comprises cabinets housing energy storage apparatuses, each containing energy storage units. The management system includes a storage apparatus filled with a cooling medium, a transmission assembly, and a control assembly. The transmission assembly comprises a main pipe connected to the storage apparatus, which branches into main and sub connecting pipes to link directly to each energy storage unit. The control assembly is mounted on the transmission assembly and partially disposed inside the energy storage unit. It automatically triggers the delivery of the cooling medium into the energy storage unit whenever the ambient temperature reaches or exceeds the unit's thermal runaway threshold.

Abstract: A data access method includes a target storage node that determines at least one data access policy input or selected by a user on a management platform. The target storage node receives a first write data stream from a request client, where the first write data stream indicates a first tape, first data to be written into the first tape, and first tape space required by the first data, the first tape space belongs to a same data partition. The target storage node writes the first data into the first tape according to the at least one data access policy using a first tape drive in at least one tape drive.

Abstract: A code generation method includes that a code development platform receives input information of a user in a first code file in a project, obtains an intra-file context and a first cross-file context based on a static structure of the project, obtains a second cross-file context based on a behavior feature of the user in development of the project, obtains a third cross-file context based on an evolutionary coupling degree between at least one second code file and the first code file in a code repository of the project, generates prompt information based on the input information, the intra-file context, the first cross-file context, the second cross-file context, and the third cross-file context, inputs the prompt information into a code generation model for inference, to obtain at least one group of generated code, and displays the at least one group of generated code to the user.

Abstract: A code testing method is applied to a code testing system. During generation of test code, context information of to-be-tested code is extracted, and the test code is generated based on the context information of the to-be-tested code.

Abstract: A data storage method includes receiving pieces of data, where each piece of data includes tag values. The tags include a first tag and a second tag that are specified as a sort key, and the first tag ranks before the second tag in the sort key. The method determines a tag value set to which a tag value of the first tag in each piece of data belongs, and determines a storage sequence of the pieces of data based on a sequence in the sort key, identification information of the tag value set to which the tag value of the first tag in each piece of data belongs, and a tag value of the second tag in each piece of data. The pieces of data are stored based on the storage sequence of the pieces of data.

Abstract: This disclosure provides a conference control method. The method is applied to a conference control system, and the conference control system includes a conference server and a conference controller. The method includes: The conference server receives a conference control request sent by the conference controller, where the conference control request is used to request to execute a target instruction related to a video conference; the conference server determines at least one hard terminal that has a pairing relationship with the conference controller, where the hard terminal is configured to implement a hardware video conference based on an embedded architecture; and the conference server processes the conference control request, to control the at least one hard terminal to execute the target instruction.

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.