Abstract: A method comprises: generating, at a first computing device, a first set of gradient values associated with a data block processed by nodes of a machine learning model, the first set of gradient values being in a first data format; determining a first shared factor from the first set of gradient values, the first shared factor being in a second data format of a lower a precision than that of the first data format; and scaling the first set of gradient values with the first shared factor, to obtain a second set of gradient values having the second data format. In addition, the method comprises sending the second set of gradient values and the first shared factor to a second computing device; and, in response to receiving a third set of gradient values and a second shared factor from the second computing device, adjusting parameters of the machine learning model.

Abstract: Embodiments of the present disclosure provide methods, devices, and computer program products for processing a task. A method of processing a task comprises: receiving, at a network device and from a set of computing devices, a set of processing results derived from processing the task by the set of computing devices; in response to receiving the set of processing results, executing a reduction operation on the set of processing results; and transmitting a result of the reduction operation to the set of computing devices. In this way, embodiments of the present disclosure can significantly reduce an amount of data exchanged among a plurality of devices processing a task in parallel, and thus reduce network latency caused by data exchange.

Abstract: A method comprises: generating, at a first computing device, a first set of gradient values associated with a data block processed by nodes of a machine learning model, the first set of gradient values being in a first data format; determining a first shared factor from the first set of gradient values, the first shared factor being in a second data format of a lower a precision than that of the first data format; and scaling the first set of gradient values with the first shared factor, to obtain a second set of gradient values having the second data format. In addition, the method comprises sending the second set of gradient values and the first shared factor to a second computing device; and, in response to receiving a third set of gradient values and a second shared factor from the second computing device, adjusting parameters of the machine learning model.

Abstract: Embodiments of the present disclosure provide methods, devices, and computer program products for processing a task. A method of processing a task comprises: receiving, at a network device and from a set of computing devices, a set of processing results derived from processing the task by the set of computing devices; in response to receiving the set of processing results, executing a reduction operation on the set of processing results; and transmitting a result of the reduction operation to the set of computing devices. In this way, embodiments of the present disclosure can significantly reduce an amount of data exchanged among a plurality of devices processing a task in parallel, and thus reduce network latency caused by data exchange.