Merge pull request #1 from PaddlePaddle/master

update from origin

paddlepalm/backbone/utils/transformer.py

@@ -24,6 +24,7 @@ import paddle.fluid as fluid
 import paddle.fluid.layers as layers
 
 from paddle.fluid.layer_helper import LayerHelper as LayerHelper
+from functools import reduce # py3
 def layer_norm(x, begin_norm_axis=1, epsilon=1e-6, param_attr=None, bias_attr=None):
     helper = LayerHelper('layer_norm', **locals())
     mean = layers.reduce_mean(x, dim=begin_norm_axis, keep_dim=True)

paddlepalm/reader/utils/reader4ernie.py

@@ -639,7 +639,8 @@ class MRCReader(BaseReader):
                  for_cn=True,
                  task_id=0,
                  doc_stride=128,
-                 max_query_length=64):
+                 max_query_length=64,
+                 remove_noanswer=True):
         self.max_seq_len = max_seq_len
         self.tokenizer = tokenization.FullTokenizer(
             vocab_file=vocab_path, do_lower_case=do_lower_case)
@@ -654,6 +655,7 @@ class MRCReader(BaseReader):
         self.max_query_length = max_query_length
         self.examples = {}
         self.features = {}
+        self.remove_noanswer = remove_noanswer
 
         if random_seed is not None:
             np.random.seed(random_seed)
@@ -758,7 +760,7 @@ class MRCReader(BaseReader):
         return cur_span_index == best_span_index
 
     def _convert_example_to_feature(self, examples, max_seq_length, tokenizer,
-                                    is_training):
+                                    is_training, remove_noanswer=True):
         features = []
         unique_id = 1000000000
 
@@ -845,6 +847,8 @@ class MRCReader(BaseReader):
                     if out_of_span:
                         start_position = 0
                         end_position = 0
+                        if remove_noanswer:
+                            continue
                     else:
                         doc_offset = len(query_tokens) + 2
                         start_position = tok_start_position - doc_start + doc_offset
@@ -958,7 +962,7 @@ class MRCReader(BaseReader):
         if not examples:
             examples = self._read_json(input_file, phase == "train")
             features = self._convert_example_to_feature(
-                examples, self.max_seq_len, self.tokenizer, phase == "train")
+                examples, self.max_seq_len, self.tokenizer, phase == "train", remove_noanswer=self.remove_noanswer)
             self.examples[phase] = examples
             self.features[phase] = features
 

paddlepalm/task_instance.py

@@ -114,8 +114,10 @@ class TaskInstance(object):
 
         conf = {}
         for k, strv in self._save_protocol.items(): 
-            exec('v={}'.format(strv))
-            conf[k] = v
+            d = None
+            v = locals()
+            exec('d={}'.format(strv), globals(), v)
+            conf[k] = v['d']
         with open(os.path.join(dirpath, '__conf__'), 'w') as writer:
             writer.write(json.dumps(conf, indent=1))
         print(self._name + ': inference model saved at ' + dirpath)

paddlepalm/tokenizer/ernie_tokenizer.py

@@ -162,10 +162,12 @@ class BasicTokenizer(object):
 
     def __init__(self, do_lower_case=True):
         """Constructs a BasicTokenizer.
+
         Args:
             do_lower_case: Whether to lower case the input.
         """
         self.do_lower_case = do_lower_case
+        self._never_lowercase = ['[UNK]', '[SEP]', '[PAD]', '[CLS]', '[MASK]']
 
     def tokenize(self, text):
         """Tokenizes a piece of text."""
@@ -183,9 +185,12 @@ class BasicTokenizer(object):
         orig_tokens = whitespace_tokenize(text)
         split_tokens = []
         for token in orig_tokens:
-            if self.do_lower_case:
+            if self.do_lower_case and token not in self._never_lowercase:
                 token = token.lower()
                 token = self._run_strip_accents(token)
+            if token in self._never_lowercase:
+                split_tokens.extend([token])
+            else:
                 split_tokens.extend(self._run_split_on_punc(token))
 
         output_tokens = whitespace_tokenize(" ".join(split_tokens))
@@ -281,14 +286,18 @@ class WordpieceTokenizer(object):
 
     def tokenize(self, text):
         """Tokenizes a piece of text into its word pieces.
+
         This uses a greedy longest-match-first algorithm to perform tokenization
         using the given vocabulary.
+
         For example:
             input = "unaffable"
             output = ["un", "##aff", "##able"]
+
         Args:
             text: A single token or whitespace separated tokens. This should have
                 already been passed through `BasicTokenizer.
+
         Returns:
             A list of wordpiece tokens.
         """

setup.py

@@ -24,8 +24,8 @@ import setuptools
 with open("README.md", "r") as fh:
     long_description = fh.read()
 setuptools.setup(
-    name="paddle-palm",
-    version="1.2",
+    name="paddlepalm",
+    version="1.0.0",
     author="PaddlePaddle",
     author_email="zhangyiming04@baidu.com",
     description="A Multi-task Learning Lib for PaddlePaddle Users.",
@@ -63,6 +63,9 @@ setuptools.setup(
             'Programming Language :: Python :: 3.6',
             'Programming Language :: Python :: 3.7',
           ],
+    install_requires = [
+        'paddlepaddle-gpu>=1.6.1'
+    ]
 )