ModelScope/modelscope/models/nlp/structbert/configuration_sbert.py

# Copyright 2021-2022 The Alibaba DAMO NLP Team Authors.
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
# 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.
""" SBERT model configuration, mainly copied from :class:`~transformers.BertConfig` """
from transformers import PretrainedConfig

from modelscope.utils import logger as logging

logger = logging.get_logger(__name__)


class SbertConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration
    of a :class:`~modelscope.models.nlp.structbert.SbertModel`.
    It is used to instantiate a SBERT model according to the specified arguments.

    Configuration objects inherit from :class:`~transformers.PretrainedConfig` and can be used to control the model
    outputs. Read the documentation from :class:`~transformers.PretrainedConfig` for more information.


    Args:
        vocab_size (:obj:`int`, `optional`, defaults to 30522):
            Vocabulary size of the BERT model. Defines the number of different tokens that can be represented by the
            :obj:`inputs_ids` passed when calling :class:`~transformers.BertModel` or
            :class:`~transformers.TFBertModel`.
        hidden_size (:obj:`int`, `optional`, defaults to 768):
            Dimensionality of the encoder layers and the pooler layer.
        num_hidden_layers (:obj:`int`, `optional`, defaults to 12):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (:obj:`int`, `optional`, defaults to 12):
            Number of attention heads for each attention layer in the Transformer encoder.
        intermediate_size (:obj:`int`, `optional`, defaults to 3072):
            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.
        hidden_act (:obj:`str` or :obj:`Callable`, `optional`, defaults to :obj:`"gelu"`):
            The non-linear activation function (function or string) in the encoder and pooler. If string,
            :obj:`"gelu"`, :obj:`"relu"`, :obj:`"silu"` and :obj:`"gelu_new"` are supported.
        hidden_dropout_prob (:obj:`float`, `optional`, defaults to 0.1):
            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
        attention_probs_dropout_prob (:obj:`float`, `optional`, defaults to 0.1):
            The dropout ratio for the attention probabilities.
        max_position_embeddings (:obj:`int`, `optional`, defaults to 512):
            The maximum sequence length that this model might ever be used with. Typically set this to something large
            just in case (e.g., 512 or 1024 or 2048).
        type_vocab_size (:obj:`int`, `optional`, defaults to 2):
            The vocabulary size of the :obj:`token_type_ids` passed when calling :class:`~transformers.BertModel` or
            :class:`~transformers.TFBertModel`.
        initializer_range (:obj:`float`, `optional`, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        layer_norm_eps (:obj:`float`, `optional`, defaults to 1e-12):
            The epsilon used by the layer normalization layers.
        position_embedding_type (:obj:`str`, `optional`, defaults to :obj:`"absolute"`):
            Type of position embedding. Choose one of :obj:`"absolute"`, :obj:`"relative_key"`,
            :obj:`"relative_key_query"`. For positional embeddings use :obj:`"absolute"`. For more information on
            :obj:`"relative_key"`, please refer to `Self-Attention with Relative Position Representations (Shaw et al.)
            <https://arxiv.org/abs/1803.02155>`__. For more information on :obj:`"relative_key_query"`, please refer to
            `Method 4` in `Improve Transformer Models with Better Relative Position Embeddings (Huang et al.)
            <https://arxiv.org/abs/2009.13658>`__.
        use_cache (:obj:`bool`, `optional`, defaults to :obj:`True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if ``config.is_decoder=True``.
        classifier_dropout (:obj:`float`, `optional`):
            The dropout ratio for the classification head.
        adv_grad_factor (:obj:`float`, `optional`): This factor will be multipled by the KL loss grad and then
            the result will be added to the original embedding.
            More details please check:https://arxiv.org/abs/1908.04577
            The range of this value always be 1e-3~1e-7
        adv_bound (:obj:`float`, `optional`): adv_bound is used to cut the top and the bottom bound of
            the produced embedding.
            If not proveded, 2 * sigma will be used as the adv_bound factor
        sigma (:obj:`float`, `optional`): The std factor used to produce a 0 mean normal distribution.
            If adv_bound not proveded, 2 * sigma will be used as the adv_bound factor
    """

    model_type = 'sbert'

    def __init__(self,
                 vocab_size=30522,
                 hidden_size=768,
                 num_hidden_layers=12,
                 num_attention_heads=12,
                 intermediate_size=3072,
                 hidden_act='gelu',
                 hidden_dropout_prob=0.1,
                 attention_probs_dropout_prob=0.1,
                 max_position_embeddings=512,
                 type_vocab_size=2,
                 initializer_range=0.02,
                 layer_norm_eps=1e-12,
                 pad_token_id=0,
                 position_embedding_type='absolute',
                 use_cache=True,
                 classifier_dropout=None,
                 **kwargs):
        super().__init__(pad_token_id=pad_token_id, **kwargs)

        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.type_vocab_size = type_vocab_size
        self.initializer_range = initializer_range
        self.layer_norm_eps = layer_norm_eps
        self.position_embedding_type = position_embedding_type
        self.use_cache = use_cache
        self.classifier_dropout = classifier_dropout
        # adv_grad_factor, used in adv loss.
        # Users can check adv_utils.py for details.
        # if adv_grad_factor set to None, no adv loss will not applied to the model.
        self.adv_grad_factor = 5e-5 if 'adv_grad_factor' not in kwargs else kwargs[
            'adv_grad_factor']
        # sigma value, used in adv loss.
        self.sigma = 5e-6 if 'sigma' not in kwargs else kwargs['sigma']
        # adv_bound value, used in adv loss.
        self.adv_bound = 2 * self.sigma if 'adv_bound' not in kwargs else kwargs[
            'adv_bound']