Skip to content

P
Paddle

PaddlePaddle / Paddle
1 年多 前同步成功

通知 2302
Star 20931
Fork 5422
P
Paddle
开放中
Opened by saxon_zh@saxon_zhOwner

【有复现问题的代码片段】为什么经过了softmax_with_cross_entropy之后，针对loss能取梯度，但是针对logits无法取梯度呢？

Created by: OleNet

import unittest
import sys

import math
import numpy as np

import paddle
import paddle.fluid as fluid


mp, sp = fluid.Program(), fluid.Program()
with fluid.program_guard(mp, sp):
    data = fluid.layers.data(name='data', shape=[128], dtype='float32')
    label = fluid.layers.data(name='label', shape=[1], dtype='int64')
    fc = fluid.layers.fc(input=data, size=100)
    ce_loss, logits = fluid.layers.softmax_with_cross_entropy(logits=fc, label=label, return_softmax=True)

    data.stop_gradient = False

    print(fluid.backward.append_backward(ce_loss, parameter_list=[data]))
    print(fluid.backward.append_backward(logits, parameter_list=[data]))