Created by: wxl1351641822

我考虑了layers的lod_append,但他并不能改变我的seq_output 考虑使用lod_reset: 他在一般的variable上可以用，但在sequence_output上用不了 set_lod也不管用

import paddle.fluid as fluid
seq_feature=fluid.layers.lod_reset(seq_output, target_lod=[1]*128)
print(seq_feature)
name: "@HUB_bert_chinese_L-12_H-768_A-12@@HUB_bert_chinese_L-12_H-768_A-12@layer_norm_24.tmp_2"
type {
  type: LOD_TENSOR
  lod_tensor {
    tensor {
      data_type: FP32
      dims: -1
      dims: 128
      dims: 768
    }
    lod_level: 0
  }
}
persistable: false

 
import paddle.fluid as fluid
x = fluid.layers.data(name='x1', shape=[128,768], lod_level=0)
print(x)
out = fluid.layers.lod_reset(x, target_lod=[128])
print(out)


name: "x1"
type {
  type: LOD_TENSOR
  lod_tensor {
    tensor {
      data_type: FP32
      dims: -1
      dims: 128
      dims: 768
    }
    lod_level: 0
  }
}
persistable: false

name: "lod_reset_1.tmp_0"
type {
  type: LOD_TENSOR
  lod_tensor {
    tensor {
      data_type: FP32
      dims: -1
      dims: 128
      dims: 768
    }
    lod_level: 1
  }
}
persistable: false

这是我想要实现的代码：

`import paddlehub as hub
max_seq_len=128
# dataset = MyDataset()
module = hub.Module(name="bert_chinese_L-12_H-768_A-12")
inputs, outputs, program = module.context(
trainable=True, max_seq_len=max_seq_len)
print(outputs)
# Use "pooled_output" for classification tasks on an entire sentence.
pooled_output = outputs["pooled_output"]
seq_output=outputs['sequence_output']
print(seq_output)
hidden_units=[128]
cls_feats = fluid.layers.dropout(
            x=seq_output,
            dropout_prob=0,
            dropout_implementation="upscale_in_train")
print(cls_feats)
if hidden_units is not None:
    for n_hidden in hidden_units:
        # cls_feats = fluid.layers.fc(
        #     input=cls_feats, size=n_hidden, act=None)
        # print(cls_feats)
        cls_feats = fluid.layers.fc(input=cls_feats, size=n_hidden,
                    act=None, bias_attr=None)
        print(cls_feats)
        lstm_h, c = fluid.layers.dynamic_lstm(
            input=cls_feats, size=n_hidden, is_reverse=False)
        print(lstm_h)
        # # 最大池化
        lstm_max = fluid.layers.sequence_pool(input=lstm_h, pool_type='max')
        print(lstm_max)
        # # 激活函数
        cls_feats = fluid.layers.tanh(lstm_max)
        print(cls_feats)

logits = fluid.layers.fc(
    input=cls_feats,
    size=3,
    param_attr=fluid.ParamAttr(
        name="cls_out_w",
        initializer=fluid.initializer.TruncatedNormal(scale=0.02)),
    bias_attr=fluid.ParamAttr(
        name="cls_out_b", initializer=fluid.initializer.Constant(0.)),
    act="softmax")

ret_infers = fluid.layers.reshape(
    x=fluid.layers.argmax(logits, axis=1), shape=[-1, 1])
----------------------
Error Message Summary:
----------------------
Error: Size of target LoD should be greater than 1.
  [Hint: Expected level0.size() > 1UL, but received level0.size():1 <= 1UL:1.] at (/paddle/paddle/fluid/operators/lod_reset_op.h:60)
  [operator < lod_reset > error]

`