不能创建paddle.static.nn.conv2d failed

[tiger100256-hu]

bug描述 Describe the Bug

使用document https://www.paddlepaddle.org.cn/documentation/docs/en/api/paddle/static/nn/conv2d_en.html 中的实例代码 在paddlepaddle 3.0 上失败了

import paddle
paddle.enable_static()

data = paddle.static.data(name='data', shape=[None, 3, 32, 32], dtype='float32')

conv2d = paddle.static.nn.conv2d(input=data, num_filters=2, filter_size=3, act="relu")
print(conv2d.shape)

root@9a1c33cc3772:~# python test_gen.py 
Traceback (most recent call last):
  File "/root/test_gen.py", line 6, in <module>
    conv2d = paddle.static.nn.conv2d(input=data, num_filters=2, filter_size=3, act="relu")
  File "/usr/local/lib/python3.10/dist-packages/paddle/static/nn/common.py", line 1061, in conv2d
    helper.append_op(
  File "/usr/local/lib/python3.10/dist-packages/paddle/base/layer_helper.py", line 57, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/paddle/base/framework.py", line 4701, in append_op
    op = Operator(
  File "/usr/local/lib/python3.10/dist-packages/paddle/base/framework.py", line 3329, in __init__
    raise TypeError(
TypeError: The type of '%Input' in operator conv2d should be one of [str, bytes, Variable]. but received : Value(define_op_name=pd_op.data, index=
0, dtype=tensor<-1x3x32x32xf32>, stop_gradient=True)

其他补充信息 Additional Supplementary Information

ubuntu 22.04
paddlepaddle 3.0.0

[liaoxin2]

您好，根据您提供的错误信息，问题出现在paddle.static.nn.conv2d的input参数传递方式不符合新版PaddlePaddle 3.0的要求。
在PaddlePaddle 3.0及以后版本，静态图模式的API有一些变化，尤其是paddle.static.data返回的Tensor类型可能不再直接兼容conv2d的输入参数。document中的实例代码可能还没有及时更新，我来反馈一下

[tiger100256-hu]

hi 现在文档怎么样了，或者能不能给我一个ok的简单的关于static.nn.conv2d 例子，谢谢

[luotao1]

可以参考下这个PR里的例子：

• mark some deprecated api to show warning #74128

[tiger100256-hu]

谢谢，通过这个https://github.com/PaddlePaddle/Paddle/pull/74128， 如果我没有理解错误的话，在paddle 3.0.0 里面paddlepaddle计划不支持老的paddle.static, 请问哪里可以找到从paddle.static 迁移到paddle.nn 的文档吗，能否给一个例子，例如如何把如下script

paddle.enable_static()
inp_blob = np.random.randn(1, 3, 4, 4).astype(np.float32)
if paddle.__version__ >= '2.0.0':
    x = paddle.static.data(name='xxx', shape=[1, 3, 4, 4], dtype='float32')
    test_layer = paddle.static.nn.conv2d(input=x, num_filters=5, filter_size=(1, 1), stride=(1, 1), padding=(1, 1),
                                         dilation=(1, 1), groups=1, bias_attr=False)
else:
    x = fluid.data(name='xxx', shape=[1, 3, 4, 4], dtype='float32')
    test_layer = fluid.layers.conv2d(input=x, num_filters=5, filter_size=(1, 1), stride=(1, 1), padding=(1, 1),
                                 dilation=(1, 1), groups=1, bias_attr=False)
if paddle.__version__ >= '2.0.0':
    relu = paddle.nn.functional.relu(test_layer)
else:
    relu = fluid.layers.relu(test_layer)

exe = fluid.Executor(fluid.CPUPlace())
exe.run(fluid.default_startup_program())
inp_dict = {'xxx': inp_blob}
var = [relu]
res_paddle = exe.run(fluid.default_main_program(),
                     fetch_list=var, feed=inp_dict)

变成一个新的能导出paddle3.0.0 PIR的json model的例子，非常感谢，麻烦了

[tiger100256-hu]

我已经create 出例子

enable_pir = False;
if os.getenv('FLAGS_enable_pir_api') == '1':
    enable_pir = True
elif os.getenv('FLAGS_enable_pir_api') == '0':
    enable_pir = False
else:
    enable_pir = False

if paddle.__version__ >= '3.0.0' and enable_pir:
    conv_layer = paddle.nn.Conv2D(
        in_channels=3,      # Number of input channels (e.g., RGB image has 3)
        out_channels=5,    # Number of output channels (filters)
        kernel_size=1,      # Size of the convolution kernel
        stride=1,           # Stride of the convolution
        padding=1,           # Padding added to both sides of the input
        dilation=1,
        groups=1,
        bias_attr=None
    )
    net = paddle.jit.to_static(conv_layer, full_graph=True)
    net.eval()
    x = np.random.rand(1, 3, 4, 4).astype('float32');
    name = "conv2d_s"
    model_dir = os.path.join(sys.argv[1], name)
    model_path = os.path.join(model_dir, name)
    if not os.path.exists(model_dir):
        os.makedirs(model_dir)
    np.save(os.path.join(model_dir, "input0"), x)
    input_tensor = paddle.to_tensor(x)
    output = net(input_tensor)
    np.save(os.path.join(model_dir, "output0"), output.numpy())
    input_spec = [paddle.static.InputSpec(shape=[1,3,4,4], dtype='float32')]
    paddle.jit.save(net, model_path, input_spec)
    sys.exit(0)