update export

examples/model_compression/pp-minilm/README.md

@@ -189,7 +189,16 @@ sh run_clue.sh CLUEWSC2020 1e-4 32 50 128 0 ppminilm-6l-768h
 
 #### 导出微调后模型
 
-模型在训练完成之后，默认情况下参数 `--save_inference_model` 为 True，会自动保存预测模型。
+模型在训练完成之后，可以选择每个数据集下效果最好的模型进行导出：
+
+```shell
+export TASK_NAME=CLUEWSC2020
+export MODEL_PATH=ppminilm-6l-768h
+export LR=1e-4
+export BS=32
+
+python export_model.py --task_name ${TASK_NAME} --output_dir ${MODEL_PATH}/models/${TASK_NAME}/${LR}_${BS}/
+```
 
 静态图（部署）模型路径与动态图模型的路径相同，文件名为 `inference.pdmodel` , `inference.pdiparams` 和 `inference.pdiparams.info` 。
 

examples/model_compression/pp-minilm/data.py

@@ -46,6 +46,7 @@ def convert_example(example,
         label_dtype = "int64" if label_list else "float32"
         # Get the label
         example['label'] = np.array(example["label"], dtype="int64")
+        label = example['label']
     # Convert raw text to feature
     if 'keyword' in example:  # CSL
         sentence1 = " ".join(example['keyword'])
@@ -85,6 +86,6 @@ def convert_example(example,
             text_pair=example['sentence2'],
             max_seq_len=max_seq_length)
     if not is_test:
-        return example['input_ids'], example['token_type_ids'], example['label']
+        return example['input_ids'], example['token_type_ids'], label
     else:
         return example['input_ids'], example['token_type_ids']

examples/model_compression/pp-minilm/finetuning/export_model.py

@@ -0,0 +1,75 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+import argparse
+import distutils.util
+
+from paddlenlp.transformers import PPMiniLMForSequenceClassification
+
+sys.path.append("../")
+from data import METRIC_CLASSES
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+
+    # Required parameters
+    parser.add_argument(
+        "--task_name",
+        default=None,
+        type=str,
+        required=True,
+        help="The name of the task to train selected in the list: " +
+        ", ".join(METRIC_CLASSES.keys()), )
+    parser.add_argument(
+        "--output_dir",
+        default="best_clue_model",
+        type=str,
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--save_inference_model_with_tokenizer",
+        type=distutils.util.strtobool,
+        default=True,
+        help="Whether to save inference model with tokenizer.")
+
+    args = parser.parse_args()
+    return args
+
+
+def do_export(args):
+    save_path = os.path.join(args.output_dir, "inference")
+    model = PPMiniLMForSequenceClassification.from_pretrained(args.output_dir)
+    is_text_pair = True
+    if args.task_name in ('tnews', 'iflytek', 'cluewsc2020'):
+        is_text_pair = False
+    model.to_static(
+        save_path,
+        use_faster_tokenizer=args.save_inference_model_with_tokenizer,
+        is_text_pair=is_text_pair)
+
+
+def print_arguments(args):
+    """print arguments"""
+    print('-----------  Configuration Arguments -----------')
+    for arg, value in sorted(vars(args).items()):
+        print('%s: %s' % (arg, value))
+    print('------------------------------------------------')
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    print_arguments(args)
+    do_export(args)

examples/model_compression/pp-minilm/finetuning/run_clue.py

@@ -137,16 +137,6 @@ def parse_args():
         type=distutils.util.strtobool,
         default=False,
         help="Whether do train.")
-    parser.add_argument(
-        "--save_inference_model",
-        type=distutils.util.strtobool,
-        default=True,
-        help="Whether to save inference model.")
-    parser.add_argument(
-        "--save_inference_model_with_tokenizer",
-        type=distutils.util.strtobool,
-        default=True,
-        help="Whether to save inference model with tokenizer.")
     parser.add_argument(
         "--max_steps",
         default=-1,
@@ -384,18 +374,6 @@ def do_train(args):
     print("best_acc: ", best_acc)
 
 
-def export_model(args):
-    save_path = os.path.join(args.output_dir, "inference")
-    model = PPMiniLMForSequenceClassification.from_pretrained(args.output_dir)
-    is_text_pair = True
-    if args.task_name in ('tnews', 'iflytek', 'cluewsc2020'):
-        is_text_pair = False
-    model.to_static(
-        save_path,
-        use_faster_tokenizer=args.save_inference_model_with_tokenizer,
-        is_text_pair=is_text_pair)
-
-
 def print_arguments(args):
     """print arguments"""
     print('-----------  Configuration Arguments -----------')
@@ -409,7 +387,5 @@ def print_arguments(args):
     print_arguments(args)
     if args.do_train:
         do_train(args)
-        if args.save_inference_model:
-            export_model(args)
     if args.do_eval:
         do_eval(args)

examples/model_compression/pp-minilm/pruning/export_model.py

@@ -21,6 +21,7 @@
 import time
 import json
 from functools import partial
+import distutils.util
 
 import numpy as np
 import paddle
@@ -132,11 +133,17 @@ def parse_args():
         type=float,
         default=1.0,
         help="depth mult you want to export")
+    parser.add_argument(
+        "--use_faster_tokenizer",
+        type=distutils.util.strtobool,
+        default=True,
+        help="Whether to use FasterTokenizer to accelerate training or further inference."
+    )
     args = parser.parse_args()
     return args
 
 
-def do_train(args):
+def do_export(args):
     paddle.set_device("gpu" if args.n_gpu else "cpu")
     args.model_type = args.model_type.lower()
     args.task_name = args.task_name.lower()
@@ -250,4 +257,4 @@ def print_arguments(args):
 if __name__ == "__main__":
     args = parse_args()
     print_arguments(args)
-    do_train(args)
+    do_export(args)

examples/model_compression/pp-minilm/quantization/quant_post.py

@@ -135,10 +135,7 @@ def batch_generator_func_using_faster_tokenizer():
                 batch_data[0].append(data['sentence1'])
                 batch_data[1].append(data['sentence2'])
                 if len(batch_data[0]) == batch_size:
-                    yield {
-                        "input_ids": batch_data[0],
-                        "token_type_ids": batch_data[1]
-                    }
+                    yield {"text": batch_data[0], "text_pair": batch_data[1]}
                     batch_data = [[], []]
 
     paddleslim.quant.quant_post_static(

paddlenlp/transformers/ppminilm/modeling.py

@@ -173,25 +173,33 @@ def to_static(self,
                     self,
                     input_spec=[
                         paddle.static.InputSpec(
-                            shape=[None], dtype=core.VarDesc.VarType.STRINGS),
-                        paddle.static.InputSpec(
-                            shape=[None], dtype=core.VarDesc.VarType.STRINGS)
+                            shape=[None],
+                            dtype=core.VarDesc.VarType.STRINGS,
+                            name="text"), paddle.static.InputSpec(
+                                shape=[None],
+                                dtype=core.VarDesc.VarType.STRINGS,
+                                name="text_pair")
                     ])
             else:
                 model = paddle.jit.to_static(
                     self,
                     input_spec=[
                         paddle.static.InputSpec(
-                            shape=[None], dtype=core.VarDesc.VarType.STRINGS)
+                            shape=[None],
+                            dtype=core.VarDesc.VarType.STRINGS,
+                            name="text")
                     ])
         else:
             model = paddle.jit.to_static(
                 self,
                 input_spec=[
                     paddle.static.InputSpec(
-                        shape=[None, None], dtype="int64"),  # input_ids
+                        shape=[None, None], dtype="int64",
+                        name="input_ids"),  # input_ids
                     paddle.static.InputSpec(
-                        shape=[None, None], dtype="int64")  # segment_ids
+                        shape=[None, None],
+                        dtype="int64",
+                        name="token_type_ids")  # segment_ids
                 ])
         paddle.jit.save(model, output_path)
         logger.info("Already save the static model to the path %s" %