mindspore/docs/api/api_python/mindspore.DynamicLossScaleManager.txt

mindspore.DynamicLossScaleManager
==================================

.. py:class:: mindspore.DynamicLossScaleManager(init_loss_scale=16777216, scale_factor=2, scale_window=2000)

    <20><>̬<EFBFBD><CCAC><EFBFBD><EFBFBD><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD>Ĺ<EFBFBD><C4B9><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̳<EFBFBD><CCB3><EFBFBD> :class:`mindspore.LossScaleManager` <20><>

    **<2A><><EFBFBD><EFBFBD><EFBFBD><EFBFBD>**

    - **init_loss_scale** (float) - <20><>ʼ<EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD>Ĭ<EFBFBD><C4AC>ֵ<EFBFBD><D6B5>2**24<32><34>
    - **scale_factor** (int) - <20>Ŵ<EFBFBD>/<2F><>С<EFBFBD><D0A1><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ĭ<EFBFBD><C4AC>ֵ<EFBFBD><D6B5>2<EFBFBD><32>
    - **scale_window** (int) - <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>step<65><70><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ĭ<EFBFBD><C4AC>ֵ<EFBFBD><D6B5>2000<30><30>

    **<2A><><EFBFBD><EFBFBD><EFBFBD><EFBFBD>**

    >>> from mindspore import Model, nn, DynamicLossScaleManager
    >>>
    >>> net = Net()
    >>> loss_scale_manager = DynamicLossScaleManager()
    >>> optim = nn.Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9)
    >>> model = Model(net, loss_scale_manager=loss_scale_manager, optimizer=optim)
    
    .. py:method:: get_drop_overflow_update()

        <20><>ֵ<EFBFBD><D6B5>ʾ<EFBFBD>Ƿ<EFBFBD><C7B7>ڷ<EFBFBD><DAB7><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ֲ<EFBFBD><D6B2><EFBFBD><EFBFBD><EFBFBD><EFBFBD>¡<EFBFBD>

        **<2A><><EFBFBD>أ<EFBFBD>**
        
        bool<6F><6C>ʼ<EFBFBD><CABC>ΪTrue<75><65>
        
    .. py:method:: get_loss_scale()

        <20><><EFBFBD>ص<EFBFBD>ǰ<EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD>

        **<2A><><EFBFBD>أ<EFBFBD>**

        float<61><74><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD>
        
    .. py:method:: get_update_cell()

        <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> :class:`mindspore.TrainOneStepWithLossScaleCell` <20>и<EFBFBD><D0B8><EFBFBD><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD> `Cell` ʵ<><CAB5><EFBFBD><EFBFBD>

        **<2A><><EFBFBD>أ<EFBFBD>**

         :class:`mindspore.DynamicLossScaleUpdateCell` ʵ<><CAB5><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ڸ<EFBFBD><DAB8><EFBFBD><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD>
        
    .. py:method:: update_loss_scale(overflow)

        <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>״̬<D7B4><CCAC><EFBFBD><EFBFBD><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>С<EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ݶȷŴ<C8B7>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD>

        **<2A><><EFBFBD><EFBFBD><EFBFBD><EFBFBD>**
        
        overflow (bool)<29><><EFBFBD><EFBFBD>ʾ<EFBFBD>Ƿ<EFBFBD><C7B7><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>