使用paddle-gpu 3.0.0.dev20241016的包，paddle.static.nn.fc的输出维度与输入不一致

[gqzhang-ai]

bug描述 Describe the Bug

使用Paddle-gpu 20241016的daily包，运行下述代码出现如下报错，使用Paddle-gpu 3.0beta则可以正常运行。

测试代码:
import os
import unittest

import numpy as np

import paddle
from paddle import base
from paddle.base import core

SEED = 2021

class TestReduceSumNet(unittest.TestCase):

def set_reduce_sum_function(self, x):
    # keep_dim = False
    return paddle.sum(x, axis=-1)

def _test(self, run_cuda=True):
    main_prog = paddle.static.Program()
    startup_prog = paddle.static.Program()
    main_prog.random_seed = SEED
    startup_prog.random_seed = SEED
    np.random.seed(SEED)

    a_np = np.random.random(size=(2, 3, 4)).astype('float32')
    b_np = np.random.random(size=(2, 3, 4)).astype('float32')
    label_np = np.random.randint(2, size=(2, 1)).astype('int64')

    with paddle.static.program_guard(main_prog, startup_prog):
        a = paddle.static.data(name="a", shape=[2, 3, 4], dtype='float32')
        b = paddle.static.data(name="b", shape=[2, 3, 4], dtype='float32')
        label = paddle.static.data(name="label",
                                   shape=[2, 1],
                                   dtype='int64')

        a_1 = paddle.static.nn.fc(x=a,
                                  size=4,
                                  num_flatten_dims=2,
                                  activation=None)
        b_1 = paddle.static.nn.fc(x=b,
                                  size=4,
                                  num_flatten_dims=2,
                                  activation=None)
        z = paddle.add(a_1, b_1)
        z_1 = self.set_reduce_sum_function(z)

        prediction = paddle.static.nn.fc(x=z_1,
                                         size=2,
                                         activation='softmax')

        cost = paddle.nn.functional.cross_entropy(input=prediction,
                                                  label=label)
        loss = paddle.mean(cost)
        sgd = paddle.optimizer.Momentum(learning_rate=0.01)
        sgd.minimize(loss)

    if run_cuda:
        place = core.CUDAPlace(0)
    else:
        place = paddle.CPUPlace()

    exe = paddle.static.Executor(place)
    exe.run(startup_prog)

    print("Start run on {}".format(place))
    for epoch in range(100):

        pred_res, loss_res = exe.run(main_prog,
                                     feed={
                                         "a": a_np,
                                         "b": b_np,
                                         "label": label_np
                                     },
                                     fetch_list=[prediction, loss])
        if epoch % 10 == 0:
            print("Epoch {} | Prediction[0]: {}, Loss: {}".format(
                epoch, pred_res[0], loss_res))

    return pred_res, loss_res

def test_cuda(self):
    cuda_pred, cuda_loss = self._test(True)
    cpu_pred, cpu_loss = self._test(False)

    self.assertTrue(np.allclose(cuda_pred, cpu_pred))
    self.assertTrue(np.allclose(cuda_loss, cpu_loss))

if name == 'main':
paddle.enable_static()
unittest.main()

其他补充信息 Additional Supplementary Information

No response

[zhangbo9674]

这是框架的一个 bug，已提交修复 PR：https://github.com/PaddlePaddle/Paddle/pull/68806/files