Created by: SunnyGJing

做序列标注任务。 修改了ERNIE的sequence_label.py文件，把模型ernie＋fc＋softmax换成ernie＋fc＋crf 结果报错了，如下： （求解决方案，谢谢）

C++ Callstacks: Enforce failed. Expected emission_weights->NumLevels() == 1UL, but received emission_weights->NumLevels():0 != 1UL:1. The Input(Emission) should be a sequence. at [/paddle/paddle/fluid/operators/crf_decoding_op.h:39]

我的代码：

enc_out = ernie.get_sequence_output()

feature_out = fluid.layers.fc( input=enc_out, size=args.num_labels, num_flatten_dims=2, 
                  param_attr=fluid.ParamAttr( name="cls_seq_label_out_w",         
                                initializer=fluid.initializer.TruncatedNormal(scale=0.02)), 
                 bias_attr=fluid.ParamAttr( name="cls_seq_label_out_b", initializer=fluid.initializer.Constant(0.)))

feature_out = fluid.layers.flatten(feature_out, axis=2)
ret_labels = fluid.layers.flatten(labels, axis=2)
crf_cost = fluid.layers.linear_chain_crf( input=feature_out, label=ret_labels, 
                                  param_attr=fluid.ParamAttr(name='crfw', learning_rate=1e-3))
loss = fluid.layers.mean(x=crf_cost)

crf_decode = fluid.layers.crf_decoding( input=feature_out, param_attr=fluid.ParamAttr(name='crfw'))
ret_infers = fluid.layers.flatten(crf_decode, axis=2)