merge opts.

inference/run_c3_infer.py

@@ -18,45 +18,19 @@
 from uer.utils.tokenizer import *
 from uer.utils.config import load_hyperparam
 from uer.model_loader import load_model
+from uer.opts import infer_opts
 from run_classifier import batch_loader
 from run_c3 import MultipleChoice, read_dataset
 
 
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--load_model_path", default=None, type=str,
-                        help="Path of the multiple choice model.")
-    parser.add_argument("--vocab_path", type=str, required=True,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--test_path", type=str,
-                        help="Path of the testset.")
-    parser.add_argument("--prediction_path", default=None, type=str,
-                        help="Path of the prediction file.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
-
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=32,
-                        help="Batch size.")
-    parser.add_argument("--seq_length", type=int, default=512,
-                        help="Sequence length.")
+    infer_opts(parser)
+
     parser.add_argument("--max_choices_num", default=4, type=int,
                         help="The maximum number of cadicate answer, shorter than this will be padded.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
-
-    # Tokenizer options.
+
     parser.add_argument("--tokenizer", choices=["bert", "char", "space"], default="bert",
                         help="Specify the tokenizer."
                              "Original Google BERT uses bert tokenizer on Chinese corpus."

inference/run_chid_infer.py

@@ -19,6 +19,7 @@
 from uer.utils.tokenizer import *
 from uer.utils.config import load_hyperparam
 from uer.model_loader import load_model
+from uer.opts import infer_opts
 from run_classifier import batch_loader
 from run_c3 import MultipleChoice
 from run_chid import read_dataset
@@ -51,52 +52,18 @@ def postprocess_chid_predictions(results):
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--load_model_path", default=None, type=str,
-                        help="Path of the multiple choice model.")
-    parser.add_argument("--vocab_path", type=str, required=True,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--test_path", type=str,
-                        help="Path of the testset.")
-    parser.add_argument("--prediction_path", default=None, type=str,
-                        help="Path of the prediction file.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
-
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=32,
-                        help="Batch size.")
-    parser.add_argument("--seq_length", type=int, default=64,
-                        help="Sequence length.")
+    infer_opts(parser)
+
     parser.add_argument("--max_choices_num", default=10, type=int,
                         help="The maximum number of cadicate answer, shorter than this will be padded.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
-
-    # Tokenizer options.
-    parser.add_argument("--tokenizer", choices=["bert", "char", "space"], default="char",
-                        help="Specify the tokenizer."
-                             "Original Google BERT uses bert tokenizer on Chinese corpus."
-                             "Char tokenizer segments sentences into characters."
-                             "Space tokenizer segments sentences into words according to space."
-                        )
 
     args = parser.parse_args()
 
     # Load the hyperparameters from the config file.
     args = load_hyperparam(args)
 
     # Build tokenizer.
-    args.tokenizer = globals()[args.tokenizer.capitalize() + "Tokenizer"](args)
+    args.tokenizer = CharTokenizer(args)
 
     # Build classification model and load parameters.
     model = MultipleChoice(args)

inference/run_classifier_infer.py

@@ -19,6 +19,7 @@
 from uer.utils.config import load_hyperparam
 from uer.utils.seed import set_seed
 from uer.model_loader import load_model
+from uer.opts import infer_opts
 from run_classifier import Classifier
 
 
@@ -69,48 +70,21 @@ def read_dataset(args, path):
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--load_model_path", default=None, type=str,
-                        help="Path of the classfier model.")
-    parser.add_argument("--vocab_path", default=None, type=str,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--test_path", type=str,
-                        help="Path of the testset.")
-    parser.add_argument("--prediction_path", default=None, type=str,
-                        help="Path of the prediction file.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
-
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=128,
-                        help="Batch size.")
-    parser.add_argument("--seq_length", type=int, default=128,
-                        help="Sequence length.")
-    parser.add_argument("--labels_num", type=int, required=True,
-                        help="Number of prediction labels.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
+    infer_opts(parser)
+
     parser.add_argument("--pooling", choices=["mean", "max", "first", "last"], default="first",
                         help="Pooling type.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
 
-    # Tokenizer options.
+    parser.add_argument("--labels_num", type=int, required=True,
+                        help="Number of prediction labels.")
+
     parser.add_argument("--tokenizer", choices=["bert", "char", "space"], default="bert",
                         help="Specify the tokenizer." 
                              "Original Google BERT uses bert tokenizer on Chinese corpus."
                              "Char tokenizer segments sentences into characters."
                              "Space tokenizer segments sentences into words according to space."
                              )
 
-    # Output options.
     parser.add_argument("--output_logits", action="store_true", help="Write logits to output file.")
     parser.add_argument("--output_prob", action="store_true", help="Write probabilities to output file.")
 

inference/run_cmrc_infer.py

@@ -17,43 +17,16 @@
 from uer.utils.tokenizer import * 
 from uer.utils.vocab import Vocab
 from uer.model_loader import load_model
+from uer.opts import infer_opts
 from run_cmrc import *
 
 
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--load_model_path", default=None, type=str,
-                        help="Path of the classfier model.")
-    parser.add_argument("--vocab_path", default=None, type=str,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--test_path", type=str,
-                        help="Path of the testset.")
-    parser.add_argument("--prediction_path", default=None, type=str,
-                        help="Path of the prediction file.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
-
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=64,
-                        help="Batch size.")
-    parser.add_argument("--seq_length", type=int, default=512,
-                        help="Sequence length.")
+    infer_opts(parser)
     parser.add_argument("--doc_stride", default=128, type=int,
                         help="When splitting up a long document into chunks, how much stride to take between chunks.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
-
 
     args = parser.parse_args()
 
@@ -121,5 +94,6 @@ def main():
 
         f.write(json.dumps(output, indent=4, ensure_ascii=False) + "\n")
 
+
 if __name__ == "__main__":
     main()

inference/run_ner_infer.py

@@ -17,6 +17,7 @@
 from uer.utils.tokenizer import *
 from uer.utils.vocab import Vocab
 from uer.model_loader import load_model
+from uer.opts import infer_opts
 from run_ner import NerTagger
 
 
@@ -59,37 +60,11 @@ def batch_loader(batch_size, src, seg):
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--load_model_path", default=None, type=str,
-                        help="Path of the NER model.")
-    parser.add_argument("--vocab_path", default=None, type=str,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--test_path", type=str,
-                        help="Path of the testset.")
-    parser.add_argument("--prediction_path", default=None, type=str,
-                        help="Path of the prediction file.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
+    infer_opts(parser)
+
     parser.add_argument("--label2id_path", type=str, required=True,
                         help="Path of the label2id file.")
 
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=128,
-                        help="Batch_size.")
-    parser.add_argument("--seq_length", default=128, type=int,
-                        help="Sequence length.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
-
     args = parser.parse_args()
 
     # Load the hyperparameters of the config file.
@@ -153,7 +128,7 @@ def main():
                 for label_id in pred[j: j+seq_length_batch[j//args.seq_length]]:
                     f.write(i2l[label_id] + " ")
                 f.write("\n")
-            
+
 
 if __name__ == "__main__":
     main()

run_chid.py

@@ -21,6 +21,7 @@
 from uer.utils.config import load_hyperparam
 from uer.utils.seed import set_seed
 from uer.model_saver import save_model
+from uer.opts import finetune_opts
 from run_c3 import MultipleChoice
 from run_classifier import build_optimizer, load_or_initialize_parameters, train_model, batch_loader, evaluate
 
@@ -128,82 +129,29 @@ def read_dataset(args, data_path, answer_path):
 def main():
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 
-    # Path options.
-    parser.add_argument("--pretrained_model_path", default=None, type=str,
-                        help="Path of the pretrained model.")
-    parser.add_argument("--output_model_path", default="./models/multichoice_model.bin", type=str,
-                        help="Path of the output model.")
-    parser.add_argument("--vocab_path", default=None, type=str,
-                        help="Path of the vocabulary file.")
-    parser.add_argument("--spm_model_path", default=None, type=str,
-                        help="Path of the sentence piece model.")
-    parser.add_argument("--train_path", type=str, required=True,
-                        help="Path of the trainset.")
+    finetune_opts(parser)
+
     parser.add_argument("--train_answer_path", type=str, required=True,
                         help="Path of the answers for trainset.")
-    parser.add_argument("--dev_path", type=str, required=True,
-                        help="Path of the devset.")
     parser.add_argument("--dev_answer_path", type=str, required=True,
                         help="Path of the answers for devset.")
-    parser.add_argument("--config_path", default="./models/bert_base_config.json", type=str,
-                        help="Path of the config file.")
-
-    # Model options.
-    parser.add_argument("--batch_size", type=int, default=8,
-                        help="Batch size.")
-    parser.add_argument("--seq_length", type=int, default=64,
-                        help="Sequence length.")
-    parser.add_argument("--embedding", choices=["bert", "word"], default="bert",
-                        help="Emebdding type.")
-    parser.add_argument("--encoder", choices=["bert", "lstm", "gru", \
-                                              "cnn", "gatedcnn", "attn", "synt", \
-                                              "rcnn", "crnn", "gpt", "bilstm"], \
-                                              default="bert", help="Encoder type.")
-    parser.add_argument("--bidirectional", action="store_true", help="Specific to recurrent model.")
-    parser.add_argument("--factorized_embedding_parameterization", action="store_true", help="Factorized embedding parameterization.")
-    parser.add_argument("--parameter_sharing", action="store_true", help="Parameter sharing.")
+
     parser.add_argument("--max_choices_num", default=10, type=int,
                         help="The maximum number of cadicate answer, shorter than this will be padded.")
 
-    # Tokenizer options.
-    parser.add_argument("--tokenizer", choices=["bert", "char", "space"], default="char",
-                        help="Specify the tokenizer."
-                             "Original Google BERT uses bert tokenizer on Chinese corpus."
-                             "Char tokenizer segments sentences into characters."
-                             "Space tokenizer segments sentences into words according to space."
-                        )
-
-    # Optimizer options.
-    parser.add_argument("--learning_rate", type=float, default=2e-5,
-                        help="Learning rate.")
-    parser.add_argument("--warmup", type=float, default=0.1,
-                        help="Warm up value.")
-    parser.add_argument("--fp16", action='store_true',
-                        help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit.")
-    parser.add_argument("--fp16_opt_level", choices=["O0", "O1", "O2", "O3" ], default='O1',
-                        help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
-                             "See details at https://nvidia.github.io/apex/amp.html")
-
-    # Training options.
-    parser.add_argument("--dropout", type=float, default=0.2,
-                        help="Dropout.")
-    parser.add_argument("--epochs_num", type=int, default=8,
-                        help="Number of epochs.")
-    parser.add_argument("--report_steps", type=int, default=100,
-                        help="Specific steps to print prompt.")
-    parser.add_argument("--seed", type=int, default=7,
-                        help="Random seed.")
-
     args = parser.parse_args()
+
     args.labels_num = args.max_choices_num
+    if args.output_model_path == None:
+        args.output_model_path = "./models/chid_model.bin"
 
     # Load the hyperparameters from the config file.
     args = load_hyperparam(args)
 
     set_seed(args.seed)
 
     # Build tokenizer.
-    args.tokenizer = globals()[args.tokenizer.capitalize() + "Tokenizer"](args)
+    args.tokenizer = CharTokenizer(args)
 
     # Build multiple choice model.
     model = MultipleChoice(args)