nthngdy commited on
Commit
b15fe97
1 Parent(s): 7fd0716

Add configuration file

Browse files

Will do the same for tokenization and modeling
Following the format of: https://huggingface.co/THUDM/chatglm-6b/tree/main

Files changed (1) hide show
  1. configuration_character_bert.py +156 -0
configuration_character_bert.py ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright Hicham EL BOUKKOURI, Olivier FERRET, Thomas LAVERGNE, Hiroshi NOJI,
3
+ # Pierre ZWEIGENBAUM, Junichi TSUJII and The HuggingFace Inc. team.
4
+ # All rights reserved.
5
+ #
6
+ # Licensed under the Apache License, Version 2.0 (the "License");
7
+ # you may not use this file except in compliance with the License.
8
+ # You may obtain a copy of the License at
9
+ #
10
+ # http://www.apache.org/licenses/LICENSE-2.0
11
+ #
12
+ # Unless required by applicable law or agreed to in writing, software
13
+ # distributed under the License is distributed on an "AS IS" BASIS,
14
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15
+ # See the License for the specific language governing permissions and
16
+ # limitations under the License.
17
+ """ CharacterBERT model configuration"""
18
+
19
+ from ...configuration_utils import PretrainedConfig
20
+ from ...utils import logging
21
+
22
+
23
+ logger = logging.get_logger(__name__)
24
+
25
+ CHARACTER_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
26
+ "helboukkouri/character-bert": "https://huggingface.co/helboukkouri/character-bert/resolve/main/config.json",
27
+ "helboukkouri/character-bert-medical": "https://huggingface.co/helboukkouri/character-bert-medical/resolve/main/config.json",
28
+ # See all CharacterBERT models at https://huggingface.co/models?filter=character_bert
29
+ }
30
+
31
+
32
+ class CharacterBertConfig(PretrainedConfig):
33
+ r"""
34
+ This is the configuration class to store the configuration of a [`CharacterBertModel`]. It is
35
+ used to instantiate an CharacterBERT model according to the specified arguments, defining the model architecture.
36
+ Instantiating a configuration with the defaults will yield a similar configuration to that of the CharacterBERT
37
+ [helboukkouri/character-bert](https://huggingface.co/helboukkouri/character-bert) architecture.
38
+
39
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model
40
+ outputs. Read the documentation from [`PretrainedConfig`] for more information.
41
+
42
+
43
+ Args:
44
+ character_embeddings_dim (`int`, *optional*, defaults to `16`):
45
+ The size of the character embeddings.
46
+ cnn_activation (`str`, *optional*, defaults to `"relu"`):
47
+ The activation function to apply to the cnn representations.
48
+ cnn_filters (:
49
+ obj:*list(list(int))*, *optional*, defaults to `[[1, 32], [2, 32], [3, 64], [4, 128], [5, 256], [6, 512], [7, 1024]]`): The list of CNN filters to use in the CharacterCNN module.
50
+ num_highway_layers (`int`, *optional*, defaults to `2`):
51
+ The number of Highway layers to apply to the CNNs output.
52
+ max_word_length (`int`, *optional*, defaults to `50`):
53
+ The maximum token length in characters (actually, in bytes as any non-ascii characters will be converted to
54
+ a sequence of utf-8 bytes).
55
+ hidden_size (`int`, *optional*, defaults to 768):
56
+ Dimensionality of the encoder layers and the pooler layer.
57
+ num_hidden_layers (`int`, *optional*, defaults to 12):
58
+ Number of hidden layers in the Transformer encoder.
59
+ num_attention_heads (`int`, *optional*, defaults to 12):
60
+ Number of attention heads for each attention layer in the Transformer encoder.
61
+ intermediate_size (`int`, *optional*, defaults to 3072):
62
+ Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
63
+ hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
64
+ The non-linear activation function (function or string) in the encoder and pooler. If string,
65
+ `"gelu"`, `"relu"`, `"selu"` and `"gelu_new"` are supported.
66
+ hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
67
+ The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
68
+ attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
69
+ The dropout ratio for the attention probabilities.
70
+ max_position_embeddings (`int`, *optional*, defaults to 512):
71
+ The maximum sequence length that this model might ever be used with. Typically set this to something large
72
+ just in case (e.g., 512 or 1024 or 2048).
73
+ type_vocab_size (`int`, *optional*, defaults to 2):
74
+ The vocabulary size of the `token_type_ids` passed when calling
75
+ [`CharacterBertModel`] or [`TFCharacterBertModel`].
76
+ mlm_vocab_size (`int`, *optional*, defaults to 100000):
77
+ Size of the output vocabulary for MLM.
78
+ initializer_range (`float`, *optional*, defaults to 0.02):
79
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
80
+ layer_norm_eps (`float`, *optional*, defaults to 1e-12):
81
+ The epsilon used by the layer normalization layers.
82
+ use_cache (`bool`, *optional*, defaults to `True`):
83
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
84
+ relevant if `config.is_decoder=True`.
85
+
86
+ Example:
87
+
88
+ ```python
89
+
90
+ ```
91
+
92
+ >>> from transformers import CharacterBertModel, CharacterBertConfig
93
+
94
+ >>> # Initializing a CharacterBERT helboukkouri/character-bert style configuration
95
+ >>> configuration = CharacterBertConfig()
96
+
97
+ >>> # Initializing a model from the helboukkouri/character-bert style configuration
98
+ >>> model = CharacterBertModel(configuration)
99
+
100
+ >>> # Accessing the model configuration
101
+ >>> configuration = model.config
102
+ """
103
+ model_type = "character_bert"
104
+
105
+ def __init__(
106
+ self,
107
+ character_embeddings_dim=16,
108
+ cnn_activation="relu",
109
+ cnn_filters=None,
110
+ num_highway_layers=2,
111
+ max_word_length=50,
112
+ hidden_size=768,
113
+ num_hidden_layers=12,
114
+ num_attention_heads=12,
115
+ intermediate_size=3072,
116
+ hidden_act="gelu",
117
+ hidden_dropout_prob=0.1,
118
+ attention_probs_dropout_prob=0.1,
119
+ max_position_embeddings=512,
120
+ type_vocab_size=2,
121
+ mlm_vocab_size=100000,
122
+ initializer_range=0.02,
123
+ layer_norm_eps=1e-12,
124
+ is_encoder_decoder=False,
125
+ use_cache=True,
126
+ **kwargs
127
+ ):
128
+ tie_word_embeddings = kwargs.pop("tie_word_embeddings", False)
129
+ if tie_word_embeddings:
130
+ raise ValueError(
131
+ "Cannot tie word embeddings in CharacterBERT. Please set " "`config.tie_word_embeddings=False`."
132
+ )
133
+ super().__init__(
134
+ type_vocab_size=type_vocab_size,
135
+ layer_norm_eps=layer_norm_eps,
136
+ use_cache=use_cache,
137
+ tie_word_embeddings=tie_word_embeddings,
138
+ **kwargs,
139
+ )
140
+ if cnn_filters is None:
141
+ cnn_filters = [[1, 32], [2, 32], [3, 64], [4, 128], [5, 256], [6, 512], [7, 1024]]
142
+ self.character_embeddings_dim = character_embeddings_dim
143
+ self.cnn_activation = cnn_activation
144
+ self.cnn_filters = cnn_filters
145
+ self.num_highway_layers = num_highway_layers
146
+ self.max_word_length = max_word_length
147
+ self.hidden_size = hidden_size
148
+ self.num_hidden_layers = num_hidden_layers
149
+ self.num_attention_heads = num_attention_heads
150
+ self.intermediate_size = intermediate_size
151
+ self.mlm_vocab_size = mlm_vocab_size
152
+ self.hidden_act = hidden_act
153
+ self.hidden_dropout_prob = hidden_dropout_prob
154
+ self.attention_probs_dropout_prob = attention_probs_dropout_prob
155
+ self.max_position_embeddings = max_position_embeddings
156
+ self.initializer_range = initializer_range