Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for managing a storage system. The method includes: based on respective task types of a plurality of tasks to be executed, allocating the plurality of tasks to a plurality of accelerator resources in a storage system for processing; at least for a first accelerator resource in the plurality of accelerator resources, determining a first polling interval based on an average task size of a first group of tasks allocated to the first accelerator resource; and scheduling the execution of the first group of tasks at the first accelerator resource at the first polling interval. The embodiments of the present disclosure can optimize the scheduling of the tasks to be executed on the plurality of accelerator resources, thereby optimizing system performance.

A method and apparatus for differentially optimizing a quality of service (QoS) includes: establishing a system model of a multi-task unloading framework; acquiring a mode for users executing a computation task, executing, according to the mode for users executing the computation task, the system model of the multi-task unloading framework; and optimizing a quality of service (QoS) on the basis of a multi-objective optimization method for a multi-agent deep reinforcement learning. According to the present invention, an unloading policy is calculated on the basis of a multi-user differentiated QoS of a multi-agent deep reinforcement learning, and with the differentiated QoS requirements among different users in a system being considered, a global unloading decision is performed according to a task performance requirement and a network resource state, and differentiated performance optimization is performed on different user requirements, thereby effectively improving a system resource utilization rate and a user service quality.

System, methods, and other embodiments described herein relate to improving scheduling of computing tasks in a mobile environment for a vehicle. In one embodiment, a method includes receiving an offloading request associated with a computing task from the vehicle, wherein the offloading request includes context information and a task descriptor related to the computing task. The method also includes scheduling the computing task to execute on a server if the context information and the task descriptor satisfy criteria for using computing resources associated with the server for the vehicle. The method also includes partitioning the computing task into subtasks if the context information satisfies the criteria. A machine learning module may decide partitions of the computing task according to the context information. The method also includes sending a scheduling signal including a scheduling message to the vehicle and the scheduling message includes scheduling information and task partition information associated with offloading the subtasks.

Some embodiments provide a compiler for optimizing the implementation of a machine-trained network (e.g., a neural network) on an integrated circuit (IC). The compiler of some embodiments receives a specification of a machine-trained network including multiple layers of computation nodes and generates a graph representing options for implementing the machine-trained network in the IC. In some embodiments, the graph includes nodes representing options for implementing each layer of the machine-trained network and edges between nodes for different layers representing different implementations that are compatible. The compiler of some embodiments is also responsible for generating instructions relating to shutting down (and waking up) memory units of cores. In some embodiments, the memory units to shutdown are determined by the compiler based on the data that is stored or will be stored in the particular memory units.

Systems and methods are provided for implementing a self-adaptive batch dataset partitioning control process which is utilized in conjunction with a distributed deep learning model training process to optimize load balancing among a set of accelerator resources. An iterative batch size tuning process is configured to determine an optimal job partition ratio for partitioning mini-batch datasets into sub-batch datasets for processing by a set of hybrid accelerator resources, wherein the sub-batch datasets are partitioned into optimal batch sizes for processing by respective accelerator resources to minimize a time for completing the deep learning model training process.

A processing device includes an interface and one or more processing circuits. The interface is to connect to a host processor. The one or more processing circuits are to receive from the host processor, via the interface, a notification specifying an operation for execution by the processing device, the operation including (i) multiple tasks that are executable by the network device, and (ii) execution dependencies among the tasks, in response to the notification, to determine a schedule for executing the tasks, the schedule complying with the execution dependencies, and to execute the operation by executing the tasks of the operation in accordance with the schedule.

Apparatuses, systems, and techniques to instantiate execution graph resources. In at least one embodiment, one or more computational resources of a parallel processing unit (PPU) to be used by one or more execution graphs are initialized prior to performance of said one or more execution graphs by said PPU.

An electronic device may map a target application to a machine learning model matched to a request of the target application among a plurality of machine learning models, may generate an inference result for sensing data corresponding to the machine learning model based on the sensing data being sensed by the at least one sensor, and may transfer the generated inference result to at least one of the target application and another application.

A method for data processing includes: receiving capability information for processing data sent by a network-side device; determining whether the capability information satisfies a preset requirement of data to be processed; and in response to the capability information satisfying the preset requirement, sending the data to be processed to the network-side device for processing.

Architectures and techniques for managing heterogeneous sets of physical GPUs. Functionality information is collected for one or more physical GPUs with a GPU device manager coupled with a heterogeneous set of physical GPUs. At least one of the physical GPUs is to be managed as multiple virtual GPUs based on the collected functionality information with the GPU device manager. Each of the physical GPUs is classified as either a single physical GPU or as one or more virtual GPUs with the device manager. Traffic representing processing jobs to be processed is received by at least a subset of the physical GPUs via a gateway programmed by a traffic manager. The GPU application to process received processing jobs scheduled by and distributed into the scheduled GPU application with a GPU scheduler communicatively coupled with the traffic manager and with the GPU device manager.