cleaning up configuration classes

examples/summarization/configuration_bertabs.py

@@ -65,7 +65,7 @@ class BertAbsConfig(PretrainedConfig):
 
     def __init__(
         self,
-        vocab_size_or_config_json_file=30522,
+        vocab_size=30522,
         max_pos=512,
         enc_layers=6,
         enc_hidden_size=512,
@@ -81,14 +81,14 @@ class BertAbsConfig(PretrainedConfig):
     ):
         super(BertAbsConfig, self).__init__(**kwargs)
 
-        if self._input_is_path_to_json(vocab_size_or_config_json_file):
-            path_to_json = vocab_size_or_config_json_file
+        if self._input_is_path_to_json(vocab_size):
+            path_to_json = vocab_size
             with open(path_to_json, "r", encoding="utf-8") as reader:
                 json_config = json.loads(reader.read())
             for key, value in json_config.items():
                 self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.vocab_size = vocab_size_or_config_json_file
+        elif isinstance(vocab_size, int):
+            self.vocab_size = vocab_size
             self.max_pos = max_pos
 
             self.enc_layers = enc_layers

templates/adding_a_new_model/configuration_xxx.py

@@ -39,7 +39,7 @@ class XxxConfig(PretrainedConfig):
 
 
         Arguments:
-            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `XxxModel`.
+            vocab_size: Vocabulary size of `inputs_ids` in `XxxModel`.
             hidden_size: Size of the encoder layers and the pooler layer.
             num_hidden_layers: Number of hidden layers in the Transformer encoder.
             num_attention_heads: Number of attention heads for each attention layer in
@@ -64,7 +64,7 @@ class XxxConfig(PretrainedConfig):
     pretrained_config_archive_map = XXX_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=50257,
+                 vocab_size=50257,
                  n_positions=1024,
                  n_ctx=1024,
                  n_embd=768,
@@ -84,7 +84,7 @@ class XxxConfig(PretrainedConfig):
                  summary_first_dropout=0.1,
                  **kwargs):
         super(XxxConfig, self).__init__(**kwargs)
-        self.vocab_size = vocab_size_or_config_json_file if isinstance(vocab_size_or_config_json_file, six.string_types) else -1
+        self.vocab_size = vocab_size if isinstance(vocab_size, six.string_types) else -1
         self.n_ctx = n_ctx
         self.n_positions = n_positions
         self.n_embd = n_embd
@@ -102,12 +102,12 @@ class XxxConfig(PretrainedConfig):
         self.summary_activation = summary_activation
         self.summary_first_dropout = summary_first_dropout
         self.summary_proj_to_labels = summary_proj_to_labels
-        if isinstance(vocab_size_or_config_json_file, six.string_types):
-            with open(vocab_size_or_config_json_file, "r", encoding="utf-8") as reader:
+        if isinstance(vocab_size, six.string_types):
+            with open(vocab_size, "r", encoding="utf-8") as reader:
                 json_config = json.loads(reader.read())
             for key, value in json_config.items():
                 self.__dict__[key] = value
-        elif not isinstance(vocab_size_or_config_json_file, int):
+        elif not isinstance(vocab_size, int):
             raise ValueError(
                 "First argument must be either a vocabulary size (int)"
                 "or the path to a pretrained model config file (str)"

templates/adding_a_new_model/tests/modeling_tf_xxx_test.py

@@ -111,7 +111,7 @@ class TFXxxModelTest(TFCommonTestCases.TFCommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = XxxConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

templates/adding_a_new_model/tests/modeling_xxx_test.py

@@ -109,7 +109,7 @@ class XxxModelTest(CommonTestCases.CommonModelTester):
                 choice_labels = ids_tensor([self.batch_size], self.num_choices)
 
             config = XxxConfig(
-                vocab_size_or_config_json_file=self.vocab_size,
+                vocab_size=self.vocab_size,
                 hidden_size=self.hidden_size,
                 num_hidden_layers=self.num_hidden_layers,
                 num_attention_heads=self.num_attention_heads,

transformers/configuration_albert.py

@@ -37,7 +37,7 @@ class AlbertConfig(PretrainedConfig):
     pretrained_config_archive_map = ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=30000,
+                 vocab_size=30000,
                  embedding_size=128,
                  hidden_size=4096,
                  num_hidden_layers=12,
@@ -83,7 +83,7 @@ class AlbertConfig(PretrainedConfig):
         """
         super(AlbertConfig, self).__init__(**kwargs)
 
-        self.vocab_size = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.embedding_size = embedding_size
         self.hidden_size = hidden_size
         self.num_hidden_layers = num_hidden_layers

transformers/configuration_bert.py

@@ -56,7 +56,7 @@ class BertConfig(PretrainedConfig):
 
 
         Arguments:
-            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `BertModel`.
+            vocab_size: Vocabulary size of `inputs_ids` in `BertModel`.
             hidden_size: Size of the encoder layers and the pooler layer.
             num_hidden_layers: Number of hidden layers in the Transformer encoder.
             num_attention_heads: Number of attention heads for each attention layer in
@@ -81,7 +81,7 @@ class BertConfig(PretrainedConfig):
     pretrained_config_archive_map = BERT_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=30522,
+                 vocab_size=30522,
                  hidden_size=768,
                  num_hidden_layers=12,
                  num_attention_heads=12,
@@ -95,14 +95,7 @@ class BertConfig(PretrainedConfig):
                  layer_norm_eps=1e-12,
                  **kwargs):
         super(BertConfig, self).__init__(**kwargs)
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.vocab_size = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.hidden_size = hidden_size
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
@@ -114,6 +107,3 @@ class BertConfig(PretrainedConfig):
         self.type_vocab_size = type_vocab_size
         self.initializer_range = initializer_range
         self.layer_norm_eps = layer_norm_eps
-        else:
-            raise ValueError("First argument must be either a vocabulary size (int)"
-                             " or the path to a pretrained model config file (str)")

transformers/configuration_ctrl.py

@@ -31,7 +31,7 @@ class CTRLConfig(PretrainedConfig):
     """Configuration class to store the configuration of a `CTRLModel`.
 
     Args:
-        vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `CTRLModel` or a configuration json file.
+        vocab_size: Vocabulary size of `inputs_ids` in `CTRLModel` or a configuration json file.
         n_positions: Number of positional embeddings.
         n_ctx: Size of the causal mask (usually same as n_positions).
         dff: Size of the inner dimension of the FFN.
@@ -52,7 +52,7 @@ class CTRLConfig(PretrainedConfig):
 
     def __init__(
         self,
-        vocab_size_or_config_json_file=246534,
+        vocab_size=246534,
         n_positions=256,
         n_ctx=256,
         n_embd=1280,
@@ -64,8 +64,6 @@ class CTRLConfig(PretrainedConfig):
         attn_pdrop=0.1,
         layer_norm_epsilon=1e-6,
         initializer_range=0.02,
-
-        num_labels=1,
         summary_type='cls_index',
         summary_use_proj=True,
         summary_activation=None,
@@ -76,7 +74,7 @@ class CTRLConfig(PretrainedConfig):
         """Constructs CTRLConfig.
 
         Args:
-            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `CTRLModel` or a configuration json file.
+            vocab_size: Vocabulary size of `inputs_ids` in `CTRLModel` or a configuration json file.
             n_positions: Number of positional embeddings.
             n_ctx: Size of the causal mask (usually same as n_positions).
             dff: Size of the inner dimension of the FFN.
@@ -94,8 +92,7 @@ class CTRLConfig(PretrainedConfig):
                 initializing all weight matrices.
         """
         super(CTRLConfig, self).__init__(**kwargs)
-
-        self.vocab_size = vocab_size_or_config_json_file if isinstance(vocab_size_or_config_json_file, int) else -1
+        self.vocab_size = vocab_size
         self.n_ctx = n_ctx
         self.n_positions = n_positions
         self.n_embd = n_embd
@@ -108,23 +105,11 @@ class CTRLConfig(PretrainedConfig):
         self.layer_norm_epsilon = layer_norm_epsilon
         self.initializer_range = initializer_range
 
-        self.num_labels = num_labels
         self.summary_type = summary_type
         self.summary_use_proj = summary_use_proj
         self.summary_activation = summary_activation
         self.summary_first_dropout = summary_first_dropout
         self.summary_proj_to_labels = summary_proj_to_labels
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding="utf-8") as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif not isinstance(vocab_size_or_config_json_file, int):
-            raise ValueError(
-                "First argument must be either a vocabulary size (int)"
-                "or the path to a pretrained model config file (str)"
-            )
 
     @property
     def max_position_embeddings(self):

transformers/configuration_distilbert.py

@@ -37,7 +37,7 @@ class DistilBertConfig(PretrainedConfig):
     pretrained_config_archive_map = DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=30522,
+                 vocab_size=30522,
                  max_position_embeddings=512,
                  sinusoidal_pos_embds=False,
                  n_layers=6,
@@ -53,15 +53,7 @@ class DistilBertConfig(PretrainedConfig):
                  seq_classif_dropout=0.2,
                  **kwargs):
         super(DistilBertConfig, self).__init__(**kwargs)
-
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.vocab_size = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.max_position_embeddings = max_position_embeddings
         self.sinusoidal_pos_embds = sinusoidal_pos_embds
         self.n_layers = n_layers
@@ -75,9 +67,7 @@ class DistilBertConfig(PretrainedConfig):
         self.tie_weights_ = tie_weights_
         self.qa_dropout = qa_dropout
         self.seq_classif_dropout = seq_classif_dropout
-        else:
-            raise ValueError("First argument must be either a vocabulary size (int)"
-                             " or the path to a pretrained model config file (str)")
+
     @property
     def hidden_size(self):
         return self.dim

transformers/configuration_gpt2.py

@@ -36,7 +36,7 @@ class GPT2Config(PretrainedConfig):
     """Configuration class to store the configuration of a `GPT2Model`.
 
     Args:
-        vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `GPT2Model` or a configuration json file.
+        vocab_size: Vocabulary size of `inputs_ids` in `GPT2Model` or a configuration json file.
         n_positions: Number of positional embeddings.
         n_ctx: Size of the causal mask (usually same as n_positions).
         n_embd: Dimensionality of the embeddings and hidden states.
@@ -56,7 +56,7 @@ class GPT2Config(PretrainedConfig):
 
     def __init__(
         self,
-        vocab_size_or_config_json_file=50257,
+        vocab_size=50257,
         n_positions=1024,
         n_ctx=1024,
         n_embd=768,
@@ -67,8 +67,6 @@ class GPT2Config(PretrainedConfig):
         attn_pdrop=0.1,
         layer_norm_epsilon=1e-5,
         initializer_range=0.02,
-
-        num_labels=1,
         summary_type='cls_index',
         summary_use_proj=True,
         summary_activation=None,
@@ -79,7 +77,7 @@ class GPT2Config(PretrainedConfig):
         """Constructs GPT2Config.
 
         Args:
-            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `GPT2Model` or a configuration json file.
+            vocab_size: Vocabulary size of `inputs_ids` in `GPT2Model` or a configuration json file.
             n_positions: Number of positional embeddings.
             n_ctx: Size of the causal mask (usually same as n_positions).
             n_embd: Dimensionality of the embeddings and hidden states.
@@ -96,15 +94,7 @@ class GPT2Config(PretrainedConfig):
                 initializing all weight matrices.
         """
         super(GPT2Config, self).__init__(**kwargs)
-
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding="utf-8") as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.vocab_size = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.n_ctx = n_ctx
         self.n_positions = n_positions
         self.n_embd = n_embd
@@ -115,18 +105,11 @@ class GPT2Config(PretrainedConfig):
         self.attn_pdrop = attn_pdrop
         self.layer_norm_epsilon = layer_norm_epsilon
         self.initializer_range = initializer_range
-
-            self.num_labels = num_labels
         self.summary_type = summary_type
         self.summary_use_proj = summary_use_proj
         self.summary_activation = summary_activation
         self.summary_first_dropout = summary_first_dropout
         self.summary_proj_to_labels = summary_proj_to_labels
-        else:
-            raise ValueError(
-                "First argument must be either a vocabulary size (int)"
-                "or the path to a pretrained model config file (str)"
-            )
 
     @property
     def max_position_embeddings(self):

transformers/configuration_openai.py

@@ -35,7 +35,7 @@ class OpenAIGPTConfig(PretrainedConfig):
     Configuration class to store the configuration of a `OpenAIGPTModel`.
 
     Args:
-        vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `OpenAIGPTModel` or a configuration json file.
+        vocab_size: Vocabulary size of `inputs_ids` in `OpenAIGPTModel` or a configuration json file.
         n_positions: Number of positional embeddings.
         n_ctx: Size of the causal mask (usually same as n_positions).
         n_embd: Dimensionality of the embeddings and hidden states.
@@ -58,7 +58,7 @@ class OpenAIGPTConfig(PretrainedConfig):
 
     def __init__(
         self,
-        vocab_size_or_config_json_file=40478,
+        vocab_size=40478,
         n_positions=512,
         n_ctx=512,
         n_embd=768,
@@ -71,8 +71,6 @@ class OpenAIGPTConfig(PretrainedConfig):
         layer_norm_epsilon=1e-5,
         initializer_range=0.02,
         predict_special_tokens=True,
-
-        num_labels=1,
         summary_type='cls_index',
         summary_use_proj=True,
         summary_activation=None,
@@ -83,15 +81,7 @@ class OpenAIGPTConfig(PretrainedConfig):
         """Constructs OpenAIGPTConfig.
         """
         super(OpenAIGPTConfig, self).__init__(**kwargs)
-
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding="utf-8") as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.vocab_size = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.n_ctx = n_ctx
         self.n_positions = n_positions
         self.n_embd = n_embd
@@ -104,18 +94,11 @@ class OpenAIGPTConfig(PretrainedConfig):
         self.layer_norm_epsilon = layer_norm_epsilon
         self.initializer_range = initializer_range
         self.predict_special_tokens = predict_special_tokens
-
-            self.num_labels = num_labels
         self.summary_type = summary_type
         self.summary_use_proj = summary_use_proj
         self.summary_activation = summary_activation
         self.summary_first_dropout = summary_first_dropout
         self.summary_proj_to_labels = summary_proj_to_labels
-        else:
-            raise ValueError(
-                "First argument must be either a vocabulary size (int)"
-                "or the path to a pretrained model config file (str)"
-            )
 
     @property
     def max_position_embeddings(self):

transformers/configuration_transfo_xl.py

@@ -34,7 +34,7 @@ class TransfoXLConfig(PretrainedConfig):
     """Configuration class to store the configuration of a `TransfoXLModel`.
 
         Args:
-            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `TransfoXLModel` or a configuration json file.
+            vocab_size: Vocabulary size of `inputs_ids` in `TransfoXLModel` or a configuration json file.
             cutoffs: cutoffs for the adaptive softmax
             d_model: Dimensionality of the model's hidden states.
             d_embed: Dimensionality of the embeddings
@@ -68,7 +68,7 @@ class TransfoXLConfig(PretrainedConfig):
     pretrained_config_archive_map = TRANSFO_XL_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=267735,
+                 vocab_size=267735,
                  cutoffs=[20000, 40000, 200000],
                  d_model=1024,
                  d_embed=1024,
@@ -100,7 +100,7 @@ class TransfoXLConfig(PretrainedConfig):
         """Constructs TransfoXLConfig.
         """
         super(TransfoXLConfig, self).__init__(**kwargs)
-        self.n_token = vocab_size_or_config_json_file if isinstance(vocab_size_or_config_json_file, int) else -1
+        self.vocab_size = vocab_size
         self.cutoffs = []
         self.cutoffs.extend(cutoffs)
         self.tie_weight = tie_weight
@@ -133,27 +133,17 @@ class TransfoXLConfig(PretrainedConfig):
         self.init_std = init_std
         self.layer_norm_epsilon = layer_norm_epsilon
 
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif not isinstance(vocab_size_or_config_json_file, int):
-            raise ValueError("First argument must be either a vocabulary size (int)"
-                             " or the path to a pretrained model config file (str)")
-
     @property
     def max_position_embeddings(self):
         return self.tgt_len + self.ext_len + self.mem_len
 
     @property
-    def vocab_size(self):
-        return self.n_token
+    def n_token(self):  # Backward compatibility
+        return self.vocab_size
 
-    @vocab_size.setter
-    def vocab_size(self, value):
-        self.n_token = value
+    @n_token.setter
+    def n_token(self, value):  # Backward compatibility
+        self.vocab_size = value
 
     @property
     def hidden_size(self):

transformers/configuration_utils.py

@@ -49,8 +49,7 @@ class PretrainedConfig(object):
     pretrained_config_archive_map = {}
 
     def __init__(self, **kwargs):
-        self.finetuning_task = kwargs.pop('finetuning_task', None)
-        self.num_labels = kwargs.pop('num_labels', 2)
+        # Attributes with defaults
         self.output_attentions = kwargs.pop('output_attentions', False)
         self.output_hidden_states = kwargs.pop('output_hidden_states', False)
         self.output_past = kwargs.pop('output_past', True)  # Not used by all models
@@ -59,6 +58,22 @@ class PretrainedConfig(object):
         self.pruned_heads = kwargs.pop('pruned_heads', {})
         self.is_decoder = kwargs.pop('is_decoder', False)
 
+        # Fine-tuning task arguments
+        self.finetuning_task = kwargs.pop('finetuning_task', None)
+        self.num_labels = kwargs.pop('num_labels', 2)
+        self.id2label = kwargs.pop('id2label', {i: 'LABEL_{}'.format(i) for i in range(self.num_labels)})
+        self.id2label = dict((int(key), value) for key, value in self.id2label.items())
+        self.label2id = kwargs.pop('label2id', dict(zip(self.id2label.values(), self.id2label.keys())))
+        self.label2id = dict((key, int(value)) for key, value in self.label2id.items())
+
+        # Additional attributes without default values
+        for key, value in kwargs.items():
+            try:
+                setattr(self, key, value)
+            except AttributeError as err:
+                logger.error("Can't set {} with value {} for {}".format(key, value, self))
+                raise err
+
     def save_pretrained(self, save_directory):
         """ Save a configuration object to the directory `save_directory`, so that it
             can be re-loaded using the :func:`~transformers.PretrainedConfig.from_pretrained` class method.
@@ -183,17 +198,15 @@ class PretrainedConfig(object):
     @classmethod
     def from_dict(cls, json_object):
         """Constructs a `Config` from a Python dictionary of parameters."""
-        config = cls(vocab_size_or_config_json_file=-1)
-        for key, value in json_object.items():
-            setattr(config, key, value)
-        return config
+        return cls(**json_object)
 
     @classmethod
     def from_json_file(cls, json_file):
         """Constructs a `Config` from a json file of parameters."""
         with open(json_file, "r", encoding='utf-8') as reader:
             text = reader.read()
-        return cls.from_dict(json.loads(text))
+        dict_obj = json.loads(text)
+        return cls(**dict_obj)
 
     def __eq__(self, other):
         return self.__dict__ == other.__dict__

transformers/configuration_xlm.py

@@ -42,7 +42,7 @@ class XLMConfig(PretrainedConfig):
     """Configuration class to store the configuration of a `XLMModel`.
 
     Args:
-        vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `XLMModel`.
+        vocab_size: Vocabulary size of `inputs_ids` in `XLMModel`.
         d_model: Size of the encoder layers and the pooler layer.
         n_layer: Number of hidden layers in the Transformer encoder.
         n_head: Number of attention heads for each attention layer in
@@ -81,7 +81,7 @@ class XLMConfig(PretrainedConfig):
     pretrained_config_archive_map = XLM_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=30145,
+                 vocab_size=30145,
                  emb_dim=2048,
                  n_layers=12,
                  n_heads=16,
@@ -103,9 +103,6 @@ class XLMConfig(PretrainedConfig):
                  unk_index=3,
                  mask_index=5,
                  is_encoder=True,
-
-                 finetuning_task=None,
-                 num_labels=2,
                  summary_type='first',
                  summary_use_proj=True,
                  summary_activation=None,
@@ -117,15 +114,7 @@ class XLMConfig(PretrainedConfig):
         """Constructs XLMConfig.
         """
         super(XLMConfig, self).__init__(**kwargs)
-
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                self.__dict__[key] = value
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.n_words = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.emb_dim = emb_dim
         self.n_layers = n_layers
         self.n_heads = n_heads
@@ -147,8 +136,6 @@ class XLMConfig(PretrainedConfig):
         self.max_position_embeddings = max_position_embeddings
         self.embed_init_std = embed_init_std
         self.init_std = init_std
-            self.finetuning_task = finetuning_task
-            self.num_labels = num_labels
         self.summary_type = summary_type
         self.summary_use_proj = summary_use_proj
         self.summary_activation = summary_activation
@@ -156,17 +143,14 @@ class XLMConfig(PretrainedConfig):
         self.summary_first_dropout = summary_first_dropout
         self.start_n_top = start_n_top
         self.end_n_top = end_n_top
-        else:
-            raise ValueError("First argument must be either a vocabulary size (int)"
-                             " or the path to a pretrained model config file (str)")
 
     @property
-    def vocab_size(self):
-        return self.n_words
+    def n_words(self):  # For backward compatibility
+        return self.vocab_size
 
-    @vocab_size.setter
-    def vocab_size(self, value):
-        self.n_words = value
+    @n_words.setter
+    def n_words(self, value):  # For backward compatibility
+        self.vocab_size = value
 
     @property
     def hidden_size(self):

transformers/configuration_xlnet.py

@@ -35,7 +35,7 @@ class XLNetConfig(PretrainedConfig):
     """Configuration class to store the configuration of a ``XLNetModel``.
 
     Args:
-        vocab_size_or_config_json_file: Vocabulary size of ``inputs_ids`` in ``XLNetModel``.
+        vocab_size: Vocabulary size of ``inputs_ids`` in ``XLNetModel``.
         d_model: Size of the encoder layers and the pooler layer.
         n_layer: Number of hidden layers in the Transformer encoder.
         n_head: Number of attention heads for each attention layer in
@@ -72,28 +72,22 @@ class XLNetConfig(PretrainedConfig):
     pretrained_config_archive_map = XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP
 
     def __init__(self,
-                 vocab_size_or_config_json_file=32000,
+                 vocab_size=32000,
                  d_model=1024,
                  n_layer=24,
                  n_head=16,
                  d_inner=4096,
-                 max_position_embeddings=512,
                  ff_activation="gelu",
                  untie_r=True,
                  attn_type="bi",
-
                  initializer_range=0.02,
                  layer_norm_eps=1e-12,
-
                  dropout=0.1,
                  mem_len=None,
                  reuse_len=None,
                  bi_data=False,
                  clamp_len=-1,
                  same_length=False,
-
-                 finetuning_task=None,
-                 num_labels=2,
                  summary_type='last',
                  summary_use_proj=True,
                  summary_activation='tanh',
@@ -104,15 +98,7 @@ class XLNetConfig(PretrainedConfig):
         """Constructs XLNetConfig.
         """
         super(XLNetConfig, self).__init__(**kwargs)
-
-        if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2
-                        and isinstance(vocab_size_or_config_json_file, unicode)):
-            with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader:
-                json_config = json.loads(reader.read())
-            for key, value in json_config.items():
-                setattr(config, key, value)
-        elif isinstance(vocab_size_or_config_json_file, int):
-            self.n_token = vocab_size_or_config_json_file
+        self.vocab_size = vocab_size
         self.d_model = d_model
         self.n_layer = n_layer
         self.n_head = n_head
@@ -133,29 +119,24 @@ class XLNetConfig(PretrainedConfig):
         self.clamp_len = clamp_len
         self.same_length = same_length
 
-            self.finetuning_task = finetuning_task
-            self.num_labels = num_labels
         self.summary_type = summary_type
         self.summary_use_proj = summary_use_proj
         self.summary_activation = summary_activation
         self.summary_last_dropout = summary_last_dropout
         self.start_n_top = start_n_top
         self.end_n_top = end_n_top
-        else:
-            raise ValueError("First argument must be either a vocabulary size (int)"
-                             " or the path to a pretrained model config file (str)")
 
     @property
     def max_position_embeddings(self):
         return -1
 
     @property
-    def vocab_size(self):
-        return self.n_token
+    def n_token(self):  # Backward compatibility
+        return self.vocab_size
 
-    @vocab_size.setter
-    def vocab_size(self, value):
-        self.n_token = value
+    @n_token.setter
+    def n_token(self, value):  # Backward compatibility
+        self.vocab_size = value
 
     @property
     def hidden_size(self):

transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py

@@ -46,7 +46,7 @@ def convert_roberta_checkpoint_to_pytorch(roberta_checkpoint_path, pytorch_dump_
     roberta = FairseqRobertaModel.from_pretrained(roberta_checkpoint_path)
     roberta.eval()  # disable dropout
     config = BertConfig(
-        vocab_size_or_config_json_file=50265,
+        vocab_size=50265,
         hidden_size=roberta.args.encoder_embed_dim,
         num_hidden_layers=roberta.args.encoder_layers,
         num_attention_heads=roberta.args.encoder_attention_heads,

transformers/modeling_gpt2.py

@@ -634,6 +634,7 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
     """
     def __init__(self, config):
         super(GPT2DoubleHeadsModel, self).__init__(config)
+        config.num_labels = 1
         self.transformer = GPT2Model(config)
         self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
         self.multiple_choice_head = SequenceSummary(config)

transformers/modeling_tf_gpt2.py

@@ -574,6 +574,7 @@ class TFGPT2DoubleHeadsModel(TFGPT2PreTrainedModel):
     """
     def __init__(self, config, *inputs, **kwargs):
         super(TFGPT2DoubleHeadsModel, self).__init__(config, *inputs, **kwargs)
+        config.num_labels = 1
         self.transformer = TFGPT2MainLayer(config, name='transformer')
         self.multiple_choice_head = TFSequenceSummary(config, initializer_range=config.initializer_range, name='multiple_choice_head')
 

transformers/modeling_tf_transfo_xl.py

@@ -353,7 +353,7 @@ class TFTransfoXLMainLayer(tf.keras.layers.Layer):
         self.output_attentions = config.output_attentions
         self.output_hidden_states = config.output_hidden_states
 
-        self.n_token = config.n_token
+        self.n_token = config.vocab_size
 
         self.d_embed = config.d_embed
         self.d_model = config.d_model
@@ -361,7 +361,7 @@ class TFTransfoXLMainLayer(tf.keras.layers.Layer):
         self.d_head = config.d_head
         self.untie_r = config.untie_r
 
-        self.word_emb = TFAdaptiveEmbedding(config.n_token, config.d_embed, config.d_model, config.cutoffs, 
+        self.word_emb = TFAdaptiveEmbedding(config.vocab_size, config.d_embed, config.d_model, config.cutoffs, 
                                             div_val=config.div_val, init_std=config.init_std, name='word_emb')
 
         self.drop = tf.keras.layers.Dropout(config.dropout)
@@ -729,7 +729,7 @@ class TFTransfoXLLMHeadModel(TFTransfoXLPreTrainedModel):
             raise NotImplementedError
         # use adaptive softmax (including standard softmax)
         else:
-            self.crit = TFAdaptiveSoftmaxMask(config.n_token, config.d_embed, config.d_model, 
+            self.crit = TFAdaptiveSoftmaxMask(config.vocab_size, config.d_embed, config.d_model, 
                                               config.cutoffs, div_val=config.div_val, name='crit')
 
     def reset_length(self, tgt_len, ext_len, mem_len):

transformers/modeling_tf_transfo_xl_utilities.py

@@ -25,15 +25,15 @@ import tensorflow as tf
 from .modeling_tf_utils import shape_list
 
 class TFAdaptiveSoftmaxMask(tf.keras.layers.Layer):
-    def __init__(self, n_token, d_embed, d_proj, cutoffs, div_val=1,
+    def __init__(self, vocab_size, d_embed, d_proj, cutoffs, div_val=1,
                  keep_order=False, **kwargs):
         super(TFAdaptiveSoftmaxMask, self).__init__(**kwargs)
 
-        self.n_token = n_token
+        self.vocab_size = vocab_size
         self.d_embed = d_embed
         self.d_proj = d_proj
 
-        self.cutoffs = cutoffs + [n_token]
+        self.cutoffs = cutoffs + [vocab_size]
         self.cutoff_ends = [0] + self.cutoffs
         self.div_val = div_val
 
@@ -66,11 +66,11 @@ class TFAdaptiveSoftmaxMask(tf.keras.layers.Layer):
                     self.out_projs.append(weight)
                 else:
                     self.out_projs.append(None)
-                weight = self.add_weight(shape=(self.n_token, self.d_embed,),
+                weight = self.add_weight(shape=(self.vocab_size, self.d_embed,),
                                          initializer='zeros',
                                          trainable=True,
                                          name='out_layers_._{}_._weight'.format(i))
-                bias = self.add_weight(shape=(self.n_token,),
+                bias = self.add_weight(shape=(self.vocab_size,),
                                          initializer='zeros',
                                          trainable=True,
                                          name='out_layers_._{}_._bias'.format(i))
@@ -114,7 +114,7 @@ class TFAdaptiveSoftmaxMask(tf.keras.layers.Layer):
         hidden, target = inputs
         head_logprob = 0
         if self.n_clusters == 0:
-            softmax_b = tf.get_variable('bias', [n_token], initializer=tf.zeros_initializer())
+            softmax_b = tf.get_variable('bias', [self.config.vocab_size], initializer=tf.zeros_initializer())
             output = self._logit(hidden, self.out_layers[0][0], self.out_layers[0][1], self.out_projs[0])
             if target is not None:
                 loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=target, logits=output)

transformers/modeling_tf_xlnet.py

@@ -366,7 +366,7 @@ class TFXLNetMainLayer(tf.keras.layers.Layer):
         self.use_bfloat16 = config.use_bfloat16
         self.initializer_range = config.initializer_range
 
-        self.word_embedding = TFSharedEmbeddings(config.n_token, config.d_model, initializer_range=config.initializer_range, name='word_embedding')
+        self.word_embedding = TFSharedEmbeddings(config.vocab_size, config.d_model, initializer_range=config.initializer_range, name='word_embedding')
         self.layer = [TFXLNetLayer(config, name='layer_._{}'.format(i)) for i in range(config.n_layer)]
         self.dropout = tf.keras.layers.Dropout(config.dropout)