Update layoutxlm (PaddlePaddle#1318)

* update layoutxlm

* update layoutxlm

Author: yingyibiao

examples/multimodal/layoutxlm/README.md

@@ -8,16 +8,20 @@
 ### 配置环境
 环境依赖
 - cv2
-- numpy
+- sentencepiece
+- yacs
 
 安装命令：
 ```shell
 pip install opencv-python
-pip install numpy
+pip install sentencepiece
+pip install yacs
 ```
 
 ### 数据准备
-处理好的XFUND中文数据集下载地址：https://bj.bcebos.com/v1/paddlenlp/datasets/XFUND.zip
+处理好的XFUND中文数据集下载地址：https://bj.bcebos.com/v1/paddlenlp/datasets/XFUND.zip 。
+
+下载并解压该数据集，解压后将数据集放置在当前目录下。
 
 ### 执行Fine-tuning
 1. ``Semantic Entity Recognition`` 任务启动Fine-tuning的方式如下：
@@ -28,7 +32,6 @@ pip install numpy
     # best metrics: {'precision': 0.8514686248331108, 'recall': 0.9354602126879354, 'f1': 0.8914904770225406}
     ```
 
-
 2. ``Relation Extraction`` 任务启动Fine-tuning的方式如下：
     ```shell
     bash run_xfun_re.sh

examples/multimodal/layoutxlm/run_xfun_re.sh

@@ -3,10 +3,10 @@ export CUDA_VISIBLE_DEVICES=0
 python ./run_xfun_re.py \
     --model_name_or_path "layoutxlm-base-uncased" \
     --max_seq_length 512 \
-    --train_data_dir "zh.train/img" \
-    --train_label_path "zh.train/xfun_normalize_train.json" \
-    --eval_data_dir "zh.val/img" \
-    --eval_label_path "zh.val/xfun_normalize_val.json" \
+    --train_data_dir "XFUND/zh_train/image" \
+    --train_label_path "XFUND/zh_train/xfun_normalize_train.json" \
+    --eval_data_dir "XFUND/zh_val/image" \
+    --eval_label_path "XFUND/zh_val/xfun_normalize_val.json" \
     --num_train_epochs 200 \
     --eval_steps 50 \
     --save_steps 500 \

examples/multimodal/layoutxlm/run_xfun_ser.sh

@@ -3,10 +3,10 @@ export CUDA_VISIBLE_DEVICES=0
 python ./run_xfun_ser.py \
     --model_name_or_path "layoutxlm-base-uncased" \
     --max_seq_length 512 \
-    --train_data_dir "zh.train/img" \
-    --train_label_path "zh.train/xfun_normalize_train.json" \
-    --eval_data_dir "zh.val/img" \
-    --eval_label_path "zh.val/xfun_normalize_val.json" \
+    --train_data_dir "XFUND/zh_train/image" \
+    --train_label_path "XFUND/zh_train/xfun_normalize_train.json" \
+    --eval_data_dir "XFUND/zh_val/image" \
+    --eval_label_path "XFUND/zh_val/xfun_normalize_val.json" \
     --num_train_epochs 200 \
     --eval_steps 10 \
     --save_steps 500 \

paddlenlp/transformers/layoutxlm/visual_backbone.py

@@ -32,8 +32,8 @@ def read_config(fp=None):
         dir_name = os.path.dirname(os.path.abspath(__file__))
         fp = os.path.join(dir_name, "visual_backbone.yaml")
     with open(fp, "r") as fin:
-        yacs = try_import("yacs")
-        cfg = yacs.config.CfgNode().load_cfg(fin)
+        yacs_config = try_import("yacs.config")
+        cfg = yacs_config.CfgNode().load_cfg(fin)
     cfg.freeze()
     return cfg