# XLM¶

## Overview¶

The XLM model was proposed in Cross-lingual Language Model Pretraining by Guillaume Lample*, Alexis Conneau*. It’s a transformer pre-trained using one of the following objectives:

• a causal language modeling (CLM) objective (next token prediction),

• a masked language modeling (MLM) objective (Bert-like), or

• a Translation Language Modeling (TLM) object (extension of Bert’s MLM to multiple language inputs)

The abstract from the paper is the following:

Recent studies have demonstrated the efficiency of generative pretraining for English natural language understanding. In this work, we extend this approach to multiple languages and show the effectiveness of cross-lingual pretraining. We propose two methods to learn cross-lingual language models (XLMs): one unsupervised that only relies on monolingual data, and one supervised that leverages parallel data with a new cross-lingual language model objective. We obtain state-of-the-art results on cross-lingual classification, unsupervised and supervised machine translation. On XNLI, our approach pushes the state of the art by an absolute gain of 4.9% accuracy. On unsupervised machine translation, we obtain 34.3 BLEU on WMT’16 German-English, improving the previous state of the art by more than 9 BLEU. On supervised machine translation, we obtain a new state of the art of 38.5 BLEU on WMT’16 Romanian-English, outperforming the previous best approach by more than 4 BLEU. Our code and pretrained models will be made publicly available.

Tips:

• XLM has many different checkpoints, which were trained using different objectives: CLM, MLM or TLM. Make sure to select the correct objective for your task (e.g. MLM checkpoints are not suitable for generation).

• XLM has multilingual checkpoints which leverage a specific lang parameter. Check out the multi-lingual page for more information.

The original code can be found here.

## XLMConfig¶

class transformers.XLMConfig(vocab_size=30145, emb_dim=2048, n_layers=12, n_heads=16, dropout=0.1, attention_dropout=0.1, gelu_activation=True, sinusoidal_embeddings=False, causal=False, asm=False, n_langs=1, use_lang_emb=True, max_position_embeddings=512, embed_init_std=0.02209708691207961, layer_norm_eps=1e-12, init_std=0.02, bos_index=0, eos_index=1, pad_index=2, unk_index=3, mask_index=5, is_encoder=True, summary_type='first', summary_use_proj=True, summary_activation=None, summary_proj_to_labels=True, summary_first_dropout=0.1, start_n_top=5, end_n_top=5, mask_token_id=0, lang_id=0, pad_token_id=2, bos_token_id=0, **kwargs)[source]

This is the configuration class to store the configuration of a XLMModel. It is used to instantiate an XLM model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the xlm-mlm-en-2048 architecture.

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

Parameters
• vocab_size (int, optional, defaults to 30145) – Vocabulary size of the XLM model. Defines the different tokens that can be represented by the inputs_ids passed to the forward method of XLMModel.

• emb_dim (int, optional, defaults to 2048) – Dimensionality of the encoder layers and the pooler layer.

• n_layer (int, optional, defaults to 12) – Number of hidden layers in the Transformer encoder.

• n_head (int, optional, defaults to 16) – Number of attention heads for each attention layer in the Transformer encoder.

• dropout (float, optional, defaults to 0.1) – The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.

• attention_dropout (float, optional, defaults to 0.1) – The dropout probability for the attention mechanism

• gelu_activation (boolean, optional, defaults to True) – The non-linear activation function (function or string) in the encoder and pooler. If set to True, “gelu” will be used instead of “relu”.

• sinusoidal_embeddings (boolean, optional, defaults to False) – Whether to use sinusoidal positional embeddings instead of absolute positional embeddings.

• causal (boolean, optional, defaults to False) – Set this to True for the model to behave in a causal manner. Causal models use a triangular attention mask in order to only attend to the left-side context instead if a bidirectional context.

• asm (boolean, optional, defaults to False) – Whether to use an adaptive log softmax projection layer instead of a linear layer for the prediction layer.

• n_langs (int, optional, defaults to 1) – The number of languages the model handles. Set to 1 for monolingual models.

• use_lang_emb (boolean, optional, defaults to True) – Whether to use language embeddings. Some models use additional language embeddings, see the multilingual models page for information on how to use them.

• max_position_embeddings (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).

• embed_init_std (float, optional, defaults to 2048^-0.5) – The standard deviation of the truncated_normal_initializer for initializing the embedding matrices.

• init_std (int, optional, defaults to 50257) – The standard deviation of the truncated_normal_initializer for initializing all weight matrices except the embedding matrices.

• layer_norm_eps (float, optional, defaults to 1e-12) – The epsilon used by the layer normalization layers.

• bos_index (int, optional, defaults to 0) – The index of the beginning of sentence token in the vocabulary.

• eos_index (int, optional, defaults to 1) – The index of the end of sentence token in the vocabulary.

• pad_index (int, optional, defaults to 2) – The index of the padding token in the vocabulary.

• unk_index (int, optional, defaults to 3) – The index of the unknown token in the vocabulary.

• mask_index (int, optional, defaults to 5) – The index of the masking token in the vocabulary.

• is_encoder (boolean, optional, defaults to True) – Whether the initialized model should be a transformer encoder or decoder as seen in Vaswani et al.

• summary_type (string, optional, defaults to “first”) –

Argument used when doing sequence summary. Used in for the multiple choice head in XLMForSequenceClassification. Is one of the following options:

• ’last’ => take the last token hidden state (like XLNet)

• ’first’ => take the first token hidden state (like Bert)

• ’mean’ => take the mean of all tokens hidden states

• ’cls_index’ => supply a Tensor of classification token position (GPT/GPT-2)

• ’attn’ => Not implemented now, use multi-head attention

• summary_use_proj (boolean, optional, defaults to True) – Argument used when doing sequence summary. Used in for the multiple choice head in XLMForSequenceClassification. Add a projection after the vector extraction

• summary_activation (string or None, optional) – Argument used when doing sequence summary. Used in for the multiple choice head in XLMForSequenceClassification. ‘tanh’ => add a tanh activation to the output, Other => no activation.

• summary_proj_to_labels (boolean, optional, defaults to True) – Argument used when doing sequence summary. Used in for the multiple choice head in XLMForSequenceClassification. If True, the projection outputs to config.num_labels classes (otherwise to hidden_size). Default: False.

• summary_first_dropout (float, optional, defaults to 0.1) – Argument used when doing sequence summary. Used in for the multiple choice head in XLMForSequenceClassification. Add a dropout before the projection and activation

• start_n_top (int, optional, defaults to 5) – Used in the SQuAD evaluation script for XLM and XLNet.

• end_n_top (int, optional, defaults to 5) – Used in the SQuAD evaluation script for XLM and XLNet.

• mask_token_id (int, optional, defaults to 0) – Model agnostic parameter to identify masked tokens when generating text in an MLM context.

• lang_id (int, optional, defaults to 1) – The ID of the language used by the model. This parameter is used when generating text in a given language.

Example:

>>> from transformers import XLMConfig, XLMModel

>>> # Initializing a XLM configuration
>>> configuration = XLMConfig()

>>> # Initializing a model from the configuration
>>> model = XLMModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config


## XLMTokenizer¶

class transformers.XLMTokenizer(vocab_file, merges_file, unk_token='<unk>', bos_token='<s>', sep_token='</s>', pad_token='<pad>', cls_token='</s>', mask_token='<special1>', additional_special_tokens=['<special0>', '<special1>', '<special2>', '<special3>', '<special4>', '<special5>', '<special6>', '<special7>', '<special8>', '<special9>'], lang2id=None, id2lang=None, do_lowercase_and_remove_accent=True, **kwargs)[source]

BPE tokenizer for XLM

• Moses preprocessing & tokenization for most supported languages

• Language specific tokenization for Chinese (Jieba), Japanese (KyTea) and Thai (PyThaiNLP)

• (optionally) lower case & normalize all inputs text

• argument special_tokens and function set_special_tokens, can be used to add additional symbols (ex: “__classify__”) to a vocabulary

• lang2id attribute maps the languages supported by the model with their ids if provided (automatically set for pretrained vocabularies)

• id2lang attributes does reverse mapping if provided (automatically set for pretrained vocabularies)

This tokenizer inherits from PreTrainedTokenizer which contains most of the methods. Users should refer to the superclass for more information regarding methods.

Parameters
• vocab_file (string) – Vocabulary file.

• merges_file (string) – Merges file.

• do_lower_case (bool, optional, defaults to True) – Whether to lowercase the input when tokenizing.

• remove_space (bool, optional, defaults to True) – Whether to strip the text when tokenizing (removing excess spaces before and after the string).

• keep_accents (bool, optional, defaults to False) – Whether to keep accents when tokenizing.

• unk_token (string, optional, defaults to “<unk>”) – The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead.

• bos_token (string, optional, defaults to “<s>”) –

The beginning of sequence token that was used during pre-training. Can be used a sequence classifier token.

Note

When building a sequence using special tokens, this is not the token that is used for the beginning of sequence. The token used is the cls_token.

• sep_token (string, optional, defaults to “</s>”) – The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for sequence classification or for a text and a question for question answering. It is also used as the last token of a sequence built with special tokens.

• pad_token (string, optional, defaults to “<pad>”) – The token used for padding, for example when batching sequences of different lengths.

• cls_token (string, optional, defaults to “</s>”) – The classifier token which is used when doing sequence classification (classification of the whole sequence instead of per-token classification). It is the first token of the sequence when built with special tokens.

• mask_token (string, optional, defaults to “<special1>”) – The token used for masking values. This is the token used when training this model with masked language modeling. This is the token which the model will try to predict.

• additional_special_tokens (List[str], optional, defaults to ["<special0>","<special1>","<special2>","<special3>","<special4>","<special5>","<special6>","<special7>","<special8>","<special9>"]) – List of additional special tokens.

• lang2id (Dict[str, int], optional) – Dictionary mapping languages string identifiers to their IDs.

• id2lang (Dict[int, str, optional) – Dictionary mapping language IDs to their string identifiers.

• do_lowercase_and_remove_accent (bool, optional, defaults to True) – Whether to lowercase and remove accents when tokenizing.

build_inputs_with_special_tokens(token_ids_0: List[int], token_ids_1: Optional[List[int]] = None) → List[int][source]

Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and adding special tokens. A XLM sequence has the following format:

• single sequence: <s> X </s>

• pair of sequences: <s> A </s> B </s>

Parameters
• token_ids_0 (List[int]) – List of IDs to which the special tokens will be added

• token_ids_1 (List[int], optional) – Optional second list of IDs for sequence pairs.

Returns

list of input IDs with the appropriate special tokens.

Return type

List[int]

create_token_type_ids_from_sequences(token_ids_0: List[int], token_ids_1: Optional[List[int]] = None) → List[int][source]

Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An XLM sequence pair mask has the following format:

0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
| first sequence    | second sequence |


if token_ids_1 is None, only returns the first portion of the mask (0s).

Parameters
• token_ids_0 (List[int]) – List of ids.

• token_ids_1 (List[int], optional) – Optional second list of IDs for sequence pairs.

Returns

List of token type IDs according to the given sequence(s).

Return type

List[int]

get_special_tokens_mask(token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False) → List[int][source]

Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding special tokens using the tokenizer prepare_for_model methods.

Parameters
• token_ids_0 (List[int]) – List of ids.

• token_ids_1 (List[int], optional) – Optional second list of IDs for sequence pairs.

• already_has_special_tokens (bool, optional, defaults to False) – Set to True if the token list is already formatted with special tokens for the model

Returns

A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.

Return type

List[int]

save_vocabulary(save_directory)[source]

Save the vocabulary and special tokens file to a directory.

Parameters

save_directory (str) – The directory in which to save the vocabulary.

Returns

Paths to the files saved.

Return type

Tuple(str)

## XLM specific outputs¶

class transformers.modeling_xlm.XLMForQuestionAnsweringOutput(loss: Optional[torch.FloatTensor] = None, start_top_log_probs: Optional[torch.FloatTensor] = None, start_top_index: Optional[torch.LongTensor] = None, end_top_log_probs: Optional[torch.FloatTensor] = None, end_top_index: Optional[torch.LongTensor] = None, cls_logits: Optional[torch.FloatTensor] = None, hidden_states: Optional[Tuple[torch.FloatTensor]] = None, attentions: Optional[Tuple[torch.FloatTensor]] = None)[source]

Base class for outputs of question answering models using a SquadHead.

Parameters
• loss (torch.FloatTensor of shape (1,), optional, returned if both start_positions and end_positions are provided) – Classification loss as the sum of start token, end token (and is_impossible if provided) classification losses.

• start_top_log_probs (torch.FloatTensor of shape (batch_size, config.start_n_top), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the top config.start_n_top start token possibilities (beam-search).

• start_top_index (torch.LongTensor of shape (batch_size, config.start_n_top), optional, returned if start_positions or end_positions is not provided) – Indices for the top config.start_n_top start token possibilities (beam-search).

• end_top_log_probs (torch.FloatTensor of shape (batch_size, config.start_n_top * config.end_n_top), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the top config.start_n_top * config.end_n_top end token possibilities (beam-search).

• end_top_index (torch.LongTensor of shape (batch_size, config.start_n_top * config.end_n_top), optional, returned if start_positions or end_positions is not provided) – Indices for the top config.start_n_top * config.end_n_top end token possibilities (beam-search).

• cls_logits (torch.FloatTensor of shape (batch_size,), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the is_impossible label of the answers.

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) –

Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) –

Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

## XLMModel¶

class transformers.XLMModel(config)[source]

The bare XLM Model transformer outputting raw hidden-states without any specific head on top.

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMModel forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

Returns

A BaseModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) – Sequence of hidden-states at the output of the last layer of the model.

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

BaseModelOutput or tuple(torch.FloatTensor)

Example:

>>> from transformers import XLMTokenizer, XLMModel
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMModel.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
>>> outputs = model(**inputs)

>>> last_hidden_states = outputs.last_hidden_state

get_input_embeddings()[source]

Returns the model’s input embeddings.

Returns

A torch module mapping vocabulary to hidden states.

Return type

nn.Module

set_input_embeddings(new_embeddings)[source]

Set model’s input embeddings

Parameters

value (nn.Module) – A module mapping vocabulary to hidden states.

class transformers.XLMWithLMHeadModel(config)[source]

The XLM Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMWithLMHeadModel forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (torch.LongTensor of shape (batch_size, sequence_length), optional) – Labels for language modeling. Note that the labels are shifted inside the model, i.e. you can set labels = input_ids Indices are selected in [-100, 0, ..., config.vocab_size] All labels set to -100 are ignored (masked), the loss is only computed for labels in [0, ..., config.vocab_size]

Returns

A MaskedLMOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) – Masked languaged modeling (MLM) loss.

• logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) – Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

MaskedLMOutput or tuple(torch.FloatTensor)

Example:

>>> import torch
>>> from transformers import XLMTokenizer, XLMWithLMHeadModel

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMWithLMHeadModel.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
>>> outputs = model(**inputs, labels=inputs["input_ids"])
>>> loss = outputs.loss
>>> logits = outputs.logits

get_output_embeddings()[source]

Returns the model’s output embeddings.

Returns

A torch module mapping hidden states to vocabulary.

Return type

nn.Module

prepare_inputs_for_generation(input_ids, **kwargs)[source]

Implement in subclasses of PreTrainedModel for custom behavior to prepare inputs in the generate method.

## XLMForSequenceClassification¶

class transformers.XLMForSequenceClassification(config)[source]

XLM Model with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks.

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMForSequenceClassification forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (torch.LongTensor of shape (batch_size,), optional) – Labels for computing the sequence classification/regression loss. Indices should be in [0, ..., config.num_labels - 1]. If config.num_labels == 1 a regression loss is computed (Mean-Square loss), If config.num_labels > 1 a classification loss is computed (Cross-Entropy).

Returns

A SequenceClassifierOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) – Classification (or regression if config.num_labels==1) loss.

• logits (torch.FloatTensor of shape (batch_size, config.num_labels)) – Classification (or regression if config.num_labels==1) scores (before SoftMax).

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

SequenceClassifierOutput or tuple(torch.FloatTensor)

Example:

>>> from transformers import XLMTokenizer, XLMForSequenceClassification
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMForSequenceClassification.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
>>> labels = torch.tensor([1]).unsqueeze(0)  # Batch size 1
>>> outputs = model(**inputs, labels=labels)
>>> loss = outputs.loss
>>> logits = outputs.logits


## XLMForMultipleChoice¶

class transformers.XLMForMultipleChoice(config, *inputs, **kwargs)[source]

XLM Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks.

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMForMultipleChoice forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (torch.Tensor of shape (batch_size,), optional) – Labels for computing the multiple choice classification loss. Indices should be in [0, ..., num_choices] where num_choices is the size of the second dimension of the input tensors. (see input_ids above)

Returns

A MultipleChoiceModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) – Classification loss.

• logits (torch.FloatTensor of shape (batch_size, num_choices)) – num_choices is the second dimension of the input tensors. (see input_ids above).

Classification scores (before SoftMax).

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

MultipleChoiceModelOutput or tuple(torch.FloatTensor)

Example:

>>> from transformers import XLMTokenizer, XLMForMultipleChoice
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMForMultipleChoice.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> prompt = "In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced."
>>> choice0 = "It is eaten with a fork and a knife."
>>> choice1 = "It is eaten while held in the hand."
>>> labels = torch.tensor(0).unsqueeze(0)  # choice0 is correct (according to Wikipedia ;)), batch size 1

>>> encoding = tokenizer([[prompt, prompt], [choice0, choice1]], return_tensors='pt', padding=True)
>>> outputs = model(**{k: v.unsqueeze(0) for k,v in encoding.items()}, labels=labels)  # batch size is 1

>>> # the linear classifier still needs to be trained
>>> loss = outputs.loss
>>> logits = outputs.logits


## XLMForTokenClassification¶

class transformers.XLMForTokenClassification(config)[source]

XLM Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, labels=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMForTokenClassification forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (torch.LongTensor of shape (batch_size, sequence_length), optional) – Labels for computing the token classification loss. Indices should be in [0, ..., config.num_labels - 1].

Returns

A TokenClassifierOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) – Classification loss.

• logits (torch.FloatTensor of shape (batch_size, sequence_length, config.num_labels)) – Classification scores (before SoftMax).

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TokenClassifierOutput or tuple(torch.FloatTensor)

Example:

>>> from transformers import XLMTokenizer, XLMForTokenClassification
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMForTokenClassification.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
>>> labels = torch.tensor([1] * inputs["input_ids"].size(1)).unsqueeze(0)  # Batch size 1

>>> outputs = model(**inputs, labels=labels)
>>> loss = outputs.loss
>>> logits = outputs.logits


class transformers.XLMForQuestionAnsweringSimple(config)[source]

XLM Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, start_positions=None, end_positions=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMForQuestionAnsweringSimple forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• start_positions (torch.LongTensor of shape (batch_size,), optional) – Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

• end_positions (torch.LongTensor of shape (batch_size,), optional) – Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

Returns

A QuestionAnsweringModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) – Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.

• start_logits (torch.FloatTensor of shape (batch_size, sequence_length,)) – Span-start scores (before SoftMax).

• end_logits (torch.FloatTensor of shape (batch_size, sequence_length,)) – Span-end scores (before SoftMax).

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

QuestionAnsweringModelOutput or tuple(torch.FloatTensor)

Example:

>>> from transformers import XLMTokenizer, XLMForQuestionAnsweringSimple
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMForQuestionAnsweringSimple.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet"
>>> inputs = tokenizer(question, text, return_tensors='pt')
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])

>>> outputs = model(**inputs, start_positions=start_positions, end_positions=end_positions)
>>> loss = outputs.loss
>>> start_scores = outputs.start_logits
>>> end_scores = outputs.end_logits


class transformers.XLMForQuestionAnswering(config)[source]

XLM Model with a beam-search span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

forward(input_ids=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, start_positions=None, end_positions=None, is_impossible=None, cls_index=None, p_mask=None, output_attentions=None, output_hidden_states=None, return_dict=None)[source]

The XLMForQuestionAnswering forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (torch.LongTensor of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (torch.LongTensor of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (torch.LongTensor of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, torch.FloatTensor], optional) – dictionary with torch.FloatTensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) –

If set to True, the model will return a ModelOutput instead of a plain tuple.

start_positions (torch.LongTensor of shape (batch_size,), optional):

Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

end_positions (torch.LongTensor of shape (batch_size,), optional):

Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

is_impossible (torch.LongTensor of shape (batch_size,), optional):

Labels whether a question has an answer or no answer (SQuAD 2.0)

cls_index (torch.LongTensor of shape (batch_size,), optional):

Labels for position (index) of the classification token to use as input for computing plausibility of the answer.

p_mask (torch.FloatTensor of shape (batch_size, sequence_length), optional):

Optional mask of tokens which can’t be in answers (e.g. [CLS], [PAD], …). 1.0 means token should be masked. 0.0 mean token is not masked.

Returns

A XLMForQuestionAnsweringOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of torch.FloatTensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (torch.FloatTensor of shape (1,), optional, returned if both start_positions and end_positions are provided) – Classification loss as the sum of start token, end token (and is_impossible if provided) classification losses.

• start_top_log_probs (torch.FloatTensor of shape (batch_size, config.start_n_top), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the top config.start_n_top start token possibilities (beam-search).

• start_top_index (torch.LongTensor of shape (batch_size, config.start_n_top), optional, returned if start_positions or end_positions is not provided) – Indices for the top config.start_n_top start token possibilities (beam-search).

• end_top_log_probs (torch.FloatTensor of shape (batch_size, config.start_n_top * config.end_n_top), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the top config.start_n_top * config.end_n_top end token possibilities (beam-search).

• end_top_index (torch.LongTensor of shape (batch_size, config.start_n_top * config.end_n_top), optional, returned if start_positions or end_positions is not provided) – Indices for the top config.start_n_top * config.end_n_top end token possibilities (beam-search).

• cls_logits (torch.FloatTensor of shape (batch_size,), optional, returned if start_positions or end_positions is not provided) – Log probabilities for the is_impossible label of the answers.

• hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of torch.FloatTensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Example:

>>> from transformers import XLMTokenizer, XLMForQuestionAnswering
>>> import torch

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMForQuestionAnswering.from_pretrained('xlm-mlm-en-2048', return_dict=True)

>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])

>>> outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
>>> loss = outputs.loss


Return type

XLMForQuestionAnsweringOutput or tuple(torch.FloatTensor)

## TFXLMModel¶

class transformers.TFXLMModel(*args, **kwargs)[source]

The bare XLM Model transformer outputing raw hidden-states without any specific head on top.

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs, **kwargs)[source]

The TFXLMModel forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

Returns

A TFBaseModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• last_hidden_state (tf.Tensor of shape (batch_size, sequence_length, hidden_size)) – Sequence of hidden-states at the output of the last layer of the model.

• hidden_states (tuple(tf.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFBaseModelOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMModel
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMModel.from_pretrained('xlm-mlm-en-2048')

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="tf")
>>> outputs = model(inputs)

>>> last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple


class transformers.TFXLMWithLMHeadModel(*args, **kwargs)[source]

The XLM Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs, **kwargs)[source]

The TFXLMWithLMHeadModel forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

Returns

A TFXLMWithLMHeadModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• logits (tf.Tensor of shape (batch_size, sequence_length, config.vocab_size)) – Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

• hidden_states (tuple(tf.Tensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFXLMWithLMHeadModelOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMWithLMHeadModel
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMWithLMHeadModel.from_pretrained('xlm-mlm-en-2048')

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="tf")
>>> outputs = model(inputs)
>>> logits = outputs[0]

get_output_embeddings()[source]

Returns the model’s output embeddings.

Returns

A torch module mapping hidden states to vocabulary.

Return type

tf.keras.layers.Layer

prepare_inputs_for_generation(inputs, **kwargs)[source]

Implement in subclasses of TFPreTrainedModel for custom behavior to prepare inputs in the generate method.

## TFXLMForSequenceClassification¶

class transformers.TFXLMForSequenceClassification(*args, **kwargs)[source]

XLM Model with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks.

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, labels=None, training=False)[source]

The TFXLMForSequenceClassification forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (tf.Tensor of shape (batch_size,), optional) – Labels for computing the sequence classification/regression loss. Indices should be in [0, ..., config.num_labels - 1]. If config.num_labels == 1 a regression loss is computed (Mean-Square loss), If config.num_labels > 1 a classification loss is computed (Cross-Entropy).

Returns

A TFSequenceClassifierOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (tf.Tensor of shape (1,), optional, returned when labels is provided) – Classification (or regression if config.num_labels==1) loss.

• logits (tf.Tensor of shape (batch_size, config.num_labels)) – Classification (or regression if config.num_labels==1) scores (before SoftMax).

• hidden_states (tuple(tf.Tensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFSequenceClassifierOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMForSequenceClassification
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMForSequenceClassification.from_pretrained('xlm-mlm-en-2048')

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="tf")
>>> inputs["labels"] = tf.reshape(tf.constant(1), (-1, 1)) # Batch size 1

>>> outputs = model(inputs)
>>> loss, logits = outputs[:2]


## TFXLMForMultipleChoice¶

class transformers.TFXLMForMultipleChoice(*args, **kwargs)[source]

XLM Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a softmax) e.g. for RocStories/SWAG tasks.

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, labels=None, training=False)[source]

The TFXLMForMultipleChoice forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (tf.Tensor of shape (batch_size,), optional) – Labels for computing the multiple choice classification loss. Indices should be in [0, ..., num_choices] where num_choices is the size of the second dimension of the input tensors. (see input_ids above)

Returns

A TFMultipleChoiceModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (tf.Tensor of shape (1,), optional, returned when labels is provided) – Classification loss.

• logits (tf.Tensor of shape (batch_size, num_choices)) – num_choices is the second dimension of the input tensors. (see input_ids above).

Classification scores (before SoftMax).

• hidden_states (tuple(tf.Tensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFMultipleChoiceModelOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMForMultipleChoice
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMForMultipleChoice.from_pretrained('xlm-mlm-en-2048')

>>> prompt = "In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced."
>>> choice0 = "It is eaten with a fork and a knife."
>>> choice1 = "It is eaten while held in the hand."

>>> encoding = tokenizer([[prompt, prompt], [choice0, choice1]], return_tensors='tf', padding=True)
>>> inputs = {k: tf.expand_dims(v, 0) for k, v in encoding.items()}
>>> outputs = model(inputs)  # batch size is 1

>>> # the linear classifier still needs to be trained
>>> logits = outputs[0]

property dummy_inputs

Dummy inputs to build the network.

Returns

tf.Tensor with dummy inputs

## TFXLMForTokenClassification¶

class transformers.TFXLMForTokenClassification(*args, **kwargs)[source]

XLM Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, labels=None, training=False)[source]

The TFXLMForTokenClassification forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• labels (tf.Tensor of shape (batch_size, sequence_length), optional) – Labels for computing the token classification loss. Indices should be in [0, ..., config.num_labels - 1].

Returns

A TFTokenClassifierOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (tf.Tensor of shape (1,), optional, returned when labels is provided) – Classification loss.

• logits (tf.Tensor of shape (batch_size, sequence_length, config.num_labels)) – Classification scores (before SoftMax).

• hidden_states (tuple(tf.Tensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFTokenClassifierOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMForTokenClassification
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMForTokenClassification.from_pretrained('xlm-mlm-en-2048')

>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="tf")
>>> input_ids = inputs["input_ids"]
>>> inputs["labels"] = tf.reshape(tf.constant([1] * tf.size(input_ids).numpy()), (-1, tf.size(input_ids))) # Batch size 1

>>> outputs = model(inputs)
>>> loss, scores = outputs[:2]


class transformers.TFXLMForQuestionAnsweringSimple(*args, **kwargs)[source]

XLM Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

Note

TF 2.0 models accepts two formats as inputs:

• having all inputs as keyword arguments (like PyTorch models), or

• having all inputs as a list, tuple or dict in the first positional arguments.

This second option is useful when using tf.keras.Model.fit() method which currently requires having all the tensors in the first argument of the model call function: model(inputs).

If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument :

• a single Tensor with input_ids only and nothing else: model(inputs_ids)

• a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: model([input_ids, attention_mask]) or model([input_ids, attention_mask, token_type_ids])

• a dictionary with one or several input Tensors associated to the input names given in the docstring: model({'input_ids': input_ids, 'token_type_ids': token_type_ids})

Parameters

config (XLMConfig) – Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

call(inputs=None, attention_mask=None, langs=None, token_type_ids=None, position_ids=None, lengths=None, cache=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, start_positions=None, end_positions=None, training=False)[source]

The TFXLMForQuestionAnsweringSimple forward method, overrides the __call__() special method.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Parameters
• input_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length)) –

Indices of input sequence tokens in the vocabulary.

Indices can be obtained using transformers.BertTokenizer. See transformers.PreTrainedTokenizer.encode() and transformers.PreTrainedTokenizer.__call__() for details.

What are input IDs?

• attention_mask (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]: 1 for tokens that are NOT MASKED, 0 for MASKED tokens.

What are attention masks?

• langs (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

A parallel sequence of tokens to be used to indicate the language of each token in the input. Indices are languages ids which can be obtained from the language names by using two conversion mappings provided in the configuration of the model (only provided for multilingual models). More precisely, the language name -> language id mapping is in model.config.lang2id (dict str -> int) and the language id -> language name mapping is model.config.id2lang (dict int -> str).

See usage examples detailed in the multilingual documentation.

• token_type_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Segment token indices to indicate first and second portions of the inputs. Indices are selected in [0, 1]: 0 corresponds to a sentence A token, 1 corresponds to a sentence B token

What are token type IDs?

• position_ids (tf.Tensor or Numpy array of shape (batch_size, sequence_length), optional) –

Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.max_position_embeddings - 1].

What are position IDs?

• lengths (tf.Tensor or Numpy array of shape (batch_size,), optional) – Length of each sentence that can be used to avoid performing attention on padding token indices. You can also use attention_mask for the same result (see above), kept here for compatbility. Indices selected in [0, ..., input_ids.size(-1)]:

• cache (Dict[str, tf.Tensor], optional) – dictionary with tf.Tensor that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see cache output below). Can be used to speed up sequential decoding. The dictionary object will be modified in-place during the forward pass to add newly computed hidden-states.

• head_mask (tf.Tensor or Numpy array of shape (num_heads,) or (num_layers, num_heads), optional) – Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]: 1 indicates the head is not masked, 0 indicates the head is masked.

• inputs_embeds (tf.Tensor or Numpy array of shape (batch_size, sequence_length, hidden_size), optional) – Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.

• output_attentions (bool, optional) – If set to True, the attentions tensors of all attention layers are returned. See attentions under returned tensors for more detail.

• output_hidden_states (bool, optional) – If set to True, the hidden states of all layers are returned. See hidden_states under returned tensors for more detail.

• return_dict (bool, optional) – If set to True, the model will return a ModelOutput instead of a plain tuple.

• start_positions (tf.Tensor of shape (batch_size,), optional) – Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

• end_positions (tf.Tensor of shape (batch_size,), optional) – Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (sequence_length). Position outside of the sequence are not taken into account for computing the loss.

Returns

A TFQuestionAnsweringModelOutput (if return_dict=True is passed or when config.return_dict=True) or a tuple of tf.Tensor comprising various elements depending on the configuration (XLMConfig) and inputs.

• loss (tf.Tensor of shape (1,), optional, returned when labels is provided) – Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.

• start_logits (tf.Tensor of shape (batch_size, sequence_length,)) – Span-start scores (before SoftMax).

• end_logits (tf.Tensor of shape (batch_size, sequence_length,)) – Span-end scores (before SoftMax).

• hidden_states (tuple(tf.Tensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) – Tuple of tf.Tensor (one for the output of the embeddings + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the initial embedding outputs.

• attentions (tuple(tf.Tensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) – Tuple of tf.Tensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Return type

TFQuestionAnsweringModelOutput or tuple(tf.Tensor)

Example:

>>> from transformers import XLMTokenizer, TFXLMForQuestionAnsweringSimple
>>> import tensorflow as tf

>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = TFXLMForQuestionAnsweringSimple.from_pretrained('xlm-mlm-en-2048')

>>> question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet"
>>> input_dict = tokenizer(question, text, return_tensors='tf')
>>> start_scores, end_scores = model(input_dict)

>>> all_tokens = tokenizer.convert_ids_to_tokens(input_dict["input_ids"].numpy()[0])
>>> answer = ' '.join(all_tokens[tf.math.argmax(start_scores, 1)[0] : tf.math.argmax(end_scores, 1)[0]+1])