diff --git a/docs/api/paddle/nn/functional/rnnt_loss_cn.rst b/docs/api/paddle/nn/functional/rnnt_loss_cn.rst index 465e7c76769..814e369c576 100644 --- a/docs/api/paddle/nn/functional/rnnt_loss_cn.rst +++ b/docs/api/paddle/nn/functional/rnnt_loss_cn.rst @@ -5,7 +5,7 @@ rnnt_loss .. py:function:: paddle.nn.functional.rnnt_loss(input, label, input_lengths, label_lengths, blank=0, fastemit_lambda=0.001, reduction='mean', name=None) -计算 RNNT loss。该接口的底层调用了第三方 [warp-transducer](https://github.com/b-flo/warp-transducer.git) 的实现。 +计算 RNNT loss。该接口的底层调用了第三方 `warp-transducer `_ 的实现。 也可以叫做 softmax with RNNT,因为 warp-transducer 库中插入了 softmax 激活函数来对输入的值进行归一化。 参数 @@ -16,7 +16,7 @@ rnnt_loss - **label_lengths** (Tensor) - 每个标签序列的长度,它应该有形状 [batch_size] 和 dtype int64。 - **blank** (int,可选) - RNN-T loss 的空白标签索引,处于半开放区间 [0,B)。数据类型必须为 int32。默认值为 0。 - **fastemit_lambda** (float,默认 0.001) - FastEmit 的正则化参数(https://arxiv.org/pdf/2010.11148.pdf)。 - - **reduction** (str,可选) - 表示如何平均损失,候选是 ``'none'``|``'mean'``|``'sum'`` 。如果 ::attr:`reduction` 是 ``'mean'``,输出将是损失的总和并除以 batch_size;如果 :attr:`reduction` 是 ``'sum'``,返回损失的总和;如果 :attr:`reduction` 为 ``'none'``,则不应用 reduction。默认是 ``'mean'``。 + - **reduction** (str,可选) - 表示如何平均损失,候选是 ``'none'`` | ``'mean'`` | ``'sum'`` 。如果 :attr:`reduction` 是 ``'mean'``,输出将是损失的总和并除以 batch_size;如果 :attr:`reduction` 是 ``'sum'``,返回损失的总和;如果 :attr:`reduction` 为 ``'none'``,则不应用 reduction。默认是 ``'mean'``。 - **name** (str,可选) - 操作名称,默认为 None。 返回