Base class for model's outputs, with potential hidden states and attentions. Args: last_hidden_state (:obj:`tf.Tensor` of shape :obj

Base class for model's outputs, with potential hidden states and attentions. Args: last_hidden_state (:obj:`torch.FloatTensor` of sh

AddedToken represents a token to be added to a Tokenizer An AddedToken can have special options defining the way it should behave.

This subclass of `argparse.ArgumentParser` uses type hints on dataclasses to generate arguments. The class is designed to play well with the

Trainer is a simple but feature-complete training and eval loop for PyTorch, optimized for 🤗 Transformers. Args: model (:class:`~tra

Base class for outputs of sentence classification models. Args: loss (:obj:`tf.Tensor` of shape :obj:`(batch_size, )`, `optional`, r

Constructs a BasicTokenizer that will run basic tokenization (punctuation splitting, lower casing, etc.). Args: do_lower_case (:obj:

Base class for outputs of sentence classification models. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, ret

Holds the output of the :meth:`~transformers.tokenization_utils_base.PreTrainedTokenizerBase.encode_plus` and :meth:`~transformers.tokenizati

Base class for outputs of multiple choice models. Args: loss (:obj:`tf.Tensor` of shape `(batch_size, )`, `optional`, returned when

Base class for outputs of question answering models. Args: loss (:obj:`tf.Tensor` of shape :obj:`(batch_size, )`, `optional`, return

Base class for outputs of token classification models. Args: loss (:obj:`tf.Tensor` of shape :obj:`(n,)`, `optional`, where n is the

Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding). Args: last_hidden_state (:

Utility class containing a conversation and its history. This class is meant to be used as an input to the :class:`~transformers.Conversation

A single training/test example for simple sequence classification. Args: guid: Unique id for the example. text_a: string. Th

TrainingArguments is the subset of the arguments we use in our example scripts **which relate to the training loop itself**. Using :clas

Retriever used to get documents from vector queries. It retrieves the documents embeddings as well as the documents contents, and it formats

Base class for masked language models outputs. Args: loss (:obj:`tf.Tensor` of shape :obj:`(n,)`, `optional`, where n is the number

this class wraps a the TFSharedEmbeddingTokens layer into a python 'no-keras-layer' class to avoid problem with weight restoring. Also it mak

Base class for masked language models outputs. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when

Implements Adam algorithm with weight decay fix as introduced in `Decoupled Weight Decay Regularization <https://arxiv.org/abs/1711.05101>`__

Quantizes the given activation. Args: activation_bit (:obj:`int`): Bitwidth for the quantized activation. act_ra

r""" This is the configuration class to store the configuration of a :class:`~transformers.BartModel`. It is used to instantiate a BART model

Base class for outputs of question answering models. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned

Base class for sequence-to-sequence language models outputs. Args: loss (:obj:`tf.Tensor` of shape :obj:`(n,)`, `optional`, where n

Base class for model encoder's outputs that also contains : pre-computed hidden states that can speed up sequential decoding. Args:

Base class for outputs of token classification models. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, return

Index grouping entries within a tensor.

`Memory` NamedTuple have a single field `bytes` and you can get a human readable str of the number of mega bytes by calling `__repr__`

r""" :class:`~transformers.DPRConfig` is the configuration class to store the configuration of a `DPRModel`. This is the configuration class

Data collator that will dynamically pad the inputs received. Args: tokenizer (:class:`~transformers.PreTrainedTokenizer` or :class:`

Base class for outputs of multiple choice models. Args: loss (:obj:`torch.FloatTensor` of shape `(1,)`, `optional`, returned when :o

Base class for model's outputs that may also contain a past key/values (to speed up sequential decoding). Args: last_hidden_state (:

r""" Holds the output of the :meth:`~transformers.SequenceFeatureExtractor.pad` and feature extractor specific ``__call__`` methods. This

The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added betwee

Base class for causal language model (or autoregressive) outputs. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `option

Base class for sequence-to-sequence language models outputs. Args: loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`,

r""" This is the configuration class to store the configuration of a :class:`~transformers.ReformerModel`. It is used to instantiate a Reforme

Base class for model encoder's outputs that also contains : pre-computed hidden states that can speed up sequential decoding. Args:

Base class for model's outputs that also contains a pooling of the last hidden states. Args: last_hidden_state (:obj:`tf.Tensor` of

Constructs a BasicTokenizer that will run basic tokenization (punctuation splitting, lower casing, etc.). Args: do_lower_case (:obj:

r""" :class:`transformers.BeamScorer` implementing standard beam search decoding. Adapted in part from `Facebook's XLM beam search code <

Quantized version of :obj:`torch.nn.LayerNorm`. Adds quantization-specific arguments on top of :obj:`torch.nn.LayerNorm`. Args:

Optimized dropout module for stabilizing the training Args: drop_prob (float): the dropout probabilities

Multi-headed attention from 'Attention Is All You Need' paper

r""" Compute a single vector summary of a sequence hidden states. Args: config (:class:`~transformers.PretrainedConfig`):

Classification Head for transformer encoders

Data collator used for language modeling. Inputs are dynamically padded to the maximum length of a batch if they are not all of the same leng

Quantized version of :obj:`torch.nn.GELU`. Adds quantization-specific arguments on top of :obj:`torch.nn.GELU`. Args: quant_mode (:o

Quantized version of :obj:`torch.nn.Softmax`. Adds quantization-specific arguments on top of :obj:`torch.nn.Softmax`. Args: outp

Runs basic tokenization with MeCab morphological parser.

Quantized version of :obj:`torch.nn.Linear`. Adds quantization-specific arguments on top of :obj:`torch.nn.Linear`. Args: weight_bit

1D-convolutional layer as defined by Radford et al. for OpenAI GPT (and also used in GPT-2). Basically works like a linear layer but the wei

Base class for model's outputs that also contains a pooling of the last hidden states. Args: last_hidden_state (:obj:`torch.FloatTen

r""" Construct a BERT tokenizer. Based on WordPiece. This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains m

This class can be used to stop generation whenever the full generated number of tokens exceeds :obj:`max_length`. Keep in mind for decoder-on

https://arxiv.org/pdf/2004.02984.pdf

A dataset that calls prepare_seq2seq_batch.

Distributed Sampler that subsamples indices sequentially, making it easier to collate all results at the end. Even though we only use this s

Optimized dropout module for stabilizing the training Args: drop_prob (float): the dropout probabilities

Fully Connected layer with on the fly adaptive mask. If needed, a score matrix is created to store the importance of each associated weight.

The product of two indices.

The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added betwee