bug描述 Describe the Bug

# 定义输入张量的形状
batch_size = 10
channels = 3
length = 32
input_tensor = paddle.randn([batch_size, channels, length])
adaptive_avg_pool = nn.AdaptiveAvgPool2D(output_size=16)
output_tensor = adaptive_avg_pool(input_tensor)
print(output_tensor.shape)

该API支持四维输入，但是没有对输入的维度进行检查，TensorFlow在同样的情况下会报错

Input 0 of layer "xxxx" is incompatible with the layer: expected ndim=4, found ndim=2. Full shape received: (None, 120)

现在的版本会
导致运行到底层的算子代码才报错，并且没有提示信息。

其他补充信息 Additional Supplementary Information

version: 2.6.1
device：GPU