MANAGING DATA PROCESSING RESOURCES
    42.
    发明申请

    公开(公告)号:US20180032373A1

    公开(公告)日:2018-02-01

    申请号:US15223394

    申请日:2016-07-29

    IPC分类号: G06F9/50 G06N99/00

    摘要: Examples relate to managing data processing resources. In one example, a computing device may: determine, for each of a plurality of data processing jobs, that the job is independent or dependent; allocate data processing resources to an independent job processing pool or a dependent job processing pool based on an initial resource share value indicating how resources are to be allocated between job processing pools; determine a first policy for scheduling data to be processed by processing resources allocated to the independent job processing pool; determine a second policy for scheduling data to be processed by processing resources allocated to the dependent job processing pool; determine an initial parallelism value that specifies a number of concurrently processing jobs; and provide a processing device with instructions to process batches of data using the allocation of data processing resources, the first policy, the second policy, and the initial parallelism value.

    Adjustable Precision for Multi-Stage Compute Processes

    公开(公告)号:US20200042287A1

    公开(公告)日:2020-02-06

    申请号:US16052218

    申请日:2018-08-01

    IPC分类号: G06F7/483 G06N3/063 G06N3/08

    摘要: Disclosed techniques provide for dynamically changing precision of a multi-stage compute process. For example, changing neural network (NN) parameters on a per-layer basis depending on properties of incoming data streams and per-layer performance of an NN among other considerations. NNs include multiple layers that may each be calculated with a different degree of accuracy and therefore, compute resource overhead (e.g., memory, processor resources, etc.). NNs are usually trained with 32-bit or 16-bit floating-point numbers. Once trained, an NN may be deployed in production. One approach to reduce compute overhead is to reduce parameter precision of NNs to 16 or 8 for deployment. The conversion to an acceptable lower precision is usually determined manually before deployment and precision levels are fixed while deployed. Disclosed techniques and implementations address automatic rather than manual determination or precision levels for different stages and dynamically adjusting precision for each stage at run-time.