Created by: levinxo

保存模型：

# baseline model
export PYTHONIOENCODING=UTF-8   # 模型输出为Unicode编码，Python2若无此设置容易报错
python3.7 inference_model.py \
		--init_checkpoint ./model_baseline \
		--inference_save_dir ./inference_model

保存为serving预测模型：

import paddle_serving_client.io as serving_io

serving_io.inference_model_to_serving('./inference_model', serving_server="serving_server", serving_client="serving_client",  model_filename='model.pdmodel', params_filename='params.pdparams')

最后生成的serving_server_conf.prototxt有错误，导致请求时返回{"result":"Request Value Error"}

feed_var {
  name: "words"
  alias_name: "words"
  is_lod_tensor: true
  feed_type: 0
  shape: -1
}
fetch_var {
  name: "crf_decoding_0.tmp_0"
  alias_name: "crf_decoding_0.tmp_0"
  is_lod_tensor: true
  fetch_type: 1             <---   这里应该是0。0代表int，1代表float
  shape: -1
}