Skip to content

xlnet

mindnlp.transformers.models.xlnet.modeling_xlnet

PyTorch XLNet model.

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetFeedForward

Bases: Cell

XLNetFeedForward is a class that represents a feed-forward neural network layer for the XLNet model. It inherits from nn.Cell and contains methods for initializing and constructing the feed-forward layer.

The init method initializes the XLNetFeedForward object with the given configuration. It sets up the layer normalization, dense layers, dropout, and activation function based on the configuration parameters.

The construct method takes an input tensor and passes it through the feed-forward layer. It applies the layer_1, activation function, dropout, layer_2, and layer normalization operations to the input tensor, and returns the output tensor after the feed-forward processing.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
class XLNetFeedForward(nn.Cell):

    """
    XLNetFeedForward is a class that represents a feed-forward neural network layer for the XLNet model.
    It inherits from nn.Cell and contains methods for initializing and constructing the feed-forward layer.

    The __init__ method initializes the XLNetFeedForward object with the given configuration.
    It sets up the layer normalization, dense layers, dropout, and activation function based on the configuration
    parameters.

    The construct method takes an input tensor and passes it through the feed-forward layer.
    It applies the layer_1, activation function, dropout, layer_2, and layer normalization operations to the input
    tensor, and returns the output tensor after the feed-forward processing.
    """
    def __init__(self, config):
        """
        Initializes an instance of the XLNetFeedForward class.

        Args:
            self: The object instance.
            config: An instance of the configuration class containing model parameters.

        Returns:
            None

        Raises:
            None

        Description:
            This method initializes the XLNetFeedForward object by setting the layer normalization, two dense layers,
            dropout rate, and activation function.

            - self.layer_norm: A LayerNorm module that normalizes the input to the dimensions of the model's hidden size.
            It takes the following parameters:

                - config.d_model: An integer representing the size of the input and output layers.
                - epsilon: A small value added to the variance to avoid division by zero. Default value is
                'config.layer_norm_eps'.

            - self.layer_1: A Dense layer that maps the input to a hidden layer. It takes the following parameters:

                - config.d_model: An integer representing the size of the input layer.
                - config.d_inner: An integer representing the size of the hidden layer.

            - self.layer_2: A Dense layer that maps the hidden layer to the output layer.
            It takes the following parameters:

                - config.d_inner: An integer representing the size of the hidden layer.
                - config.d_model: An integer representing the size of the output layer.

            - self.dropout: A Dropout layer that randomly sets elements to zero during training to prevent overfitting.
            It takes the following parameter:

                - p: The probability of an element to be zeroed. Default value is 'config.dropout'.

            - self.activation_function: The activation function used in the feed-forward layer.
            It can be either a string representing the name of the activation function or a custom activation function.
            If it is a string, it is looked up in the ACT2FN mapping, which maps activation function names to their
            corresponding functions. Otherwise, it is directly assigned to the provided activation function.

        Note:
            - The 'config' parameter should be an instance of the configuration class, which contains necessary model
            parameters.
            - The 'config.ff_activation' parameter can be either a string or a custom activation function.
        """
        super().__init__()
        self.layer_norm = nn.LayerNorm(config.d_model, epsilon=config.layer_norm_eps)
        self.layer_1 = nn.Dense(config.d_model, config.d_inner)
        self.layer_2 = nn.Dense(config.d_inner, config.d_model)
        self.dropout = nn.Dropout(p=config.dropout)
        if isinstance(config.ff_activation, str):
            self.activation_function = ACT2FN[config.ff_activation]
        else:
            self.activation_function = config.ff_activation

    def construct(self, inp):
        """
        Constructs the XLNet feed-forward layer.

        Args:
            self (XLNetFeedForward): The instance of the XLNetFeedForward class.
            inp: The input tensor to the feed-forward layer.

        Returns:
            None

        Raises:
            None

        This method applies the XLNet feed-forward layer operations on the input tensor.
        It performs the following steps:

        1. Applies layer_1 on the input tensor.
        2. Applies the activation function on the output of layer_1.
        3. Applies dropout regularization on the output of the activation function.
        4. Applies layer_2 on the output of the dropout operation.
        5. Applies dropout regularization on the output of layer_2.
        6. Adds the input tensor to the output of layer_2 and applies layer normalization.
        7. Returns the final output tensor.

        Note:
            The input tensor is expected to have the shape (batch_size, sequence_length, hidden_size).
        """
        output = inp
        output = self.layer_1(output)
        output = self.activation_function(output)
        output = self.dropout(output)
        output = self.layer_2(output)
        output = self.dropout(output)
        output = self.layer_norm(output + inp)
        return output

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetFeedForward.__init__(config)

Initializes an instance of the XLNetFeedForward class.

PARAMETER DESCRIPTION
self

The object instance.
config

An instance of the configuration class containing model parameters.
RETURNS DESCRIPTION

None
Description

This method initializes the XLNetFeedForward object by setting the layer normalization, two dense layers, dropout rate, and activation function.

  • self.layer_norm: A LayerNorm module that normalizes the input to the dimensions of the model's hidden size. It takes the following parameters:

    • config.d_model: An integer representing the size of the input and output layers.
    • epsilon: A small value added to the variance to avoid division by zero. Default value is 'config.layer_norm_eps'.

  • self.layer_1: A Dense layer that maps the input to a hidden layer. It takes the following parameters:

    • config.d_model: An integer representing the size of the input layer.
    • config.d_inner: An integer representing the size of the hidden layer.

  • self.layer_2: A Dense layer that maps the hidden layer to the output layer. It takes the following parameters:

    • config.d_inner: An integer representing the size of the hidden layer.
    • config.d_model: An integer representing the size of the output layer.

  • self.dropout: A Dropout layer that randomly sets elements to zero during training to prevent overfitting. It takes the following parameter:

    • p: The probability of an element to be zeroed. Default value is 'config.dropout'.

  • self.activation_function: The activation function used in the feed-forward layer. It can be either a string representing the name of the activation function or a custom activation function. If it is a string, it is looked up in the ACT2FN mapping, which maps activation function names to their corresponding functions. Otherwise, it is directly assigned to the provided activation function.

Note
  • The 'config' parameter should be an instance of the configuration class, which contains necessary model parameters.
  • The 'config.ff_activation' parameter can be either a string or a custom activation function.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
def __init__(self, config):
    """
    Initializes an instance of the XLNetFeedForward class.

    Args:
        self: The object instance.
        config: An instance of the configuration class containing model parameters.

    Returns:
        None

    Raises:
        None

    Description:
        This method initializes the XLNetFeedForward object by setting the layer normalization, two dense layers,
        dropout rate, and activation function.

        - self.layer_norm: A LayerNorm module that normalizes the input to the dimensions of the model's hidden size.
        It takes the following parameters:

            - config.d_model: An integer representing the size of the input and output layers.
            - epsilon: A small value added to the variance to avoid division by zero. Default value is
            'config.layer_norm_eps'.

        - self.layer_1: A Dense layer that maps the input to a hidden layer. It takes the following parameters:

            - config.d_model: An integer representing the size of the input layer.
            - config.d_inner: An integer representing the size of the hidden layer.

        - self.layer_2: A Dense layer that maps the hidden layer to the output layer.
        It takes the following parameters:

            - config.d_inner: An integer representing the size of the hidden layer.
            - config.d_model: An integer representing the size of the output layer.

        - self.dropout: A Dropout layer that randomly sets elements to zero during training to prevent overfitting.
        It takes the following parameter:

            - p: The probability of an element to be zeroed. Default value is 'config.dropout'.

        - self.activation_function: The activation function used in the feed-forward layer.
        It can be either a string representing the name of the activation function or a custom activation function.
        If it is a string, it is looked up in the ACT2FN mapping, which maps activation function names to their
        corresponding functions. Otherwise, it is directly assigned to the provided activation function.

    Note:
        - The 'config' parameter should be an instance of the configuration class, which contains necessary model
        parameters.
        - The 'config.ff_activation' parameter can be either a string or a custom activation function.
    """
    super().__init__()
    self.layer_norm = nn.LayerNorm(config.d_model, epsilon=config.layer_norm_eps)
    self.layer_1 = nn.Dense(config.d_model, config.d_inner)
    self.layer_2 = nn.Dense(config.d_inner, config.d_model)
    self.dropout = nn.Dropout(p=config.dropout)
    if isinstance(config.ff_activation, str):
        self.activation_function = ACT2FN[config.ff_activation]
    else:
        self.activation_function = config.ff_activation

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetFeedForward.construct(inp)

Constructs the XLNet feed-forward layer.

PARAMETER DESCRIPTION
self

The instance of the XLNetFeedForward class.

TYPE: XLNetFeedForward

inp

The input tensor to the feed-forward layer.
RETURNS DESCRIPTION

None

This method applies the XLNet feed-forward layer operations on the input tensor. It performs the following steps:

  1. Applies layer_1 on the input tensor.
  2. Applies the activation function on the output of layer_1.
  3. Applies dropout regularization on the output of the activation function.
  4. Applies layer_2 on the output of the dropout operation.
  5. Applies dropout regularization on the output of layer_2.
  6. Adds the input tensor to the output of layer_2 and applies layer normalization.
  7. Returns the final output tensor.
Note

The input tensor is expected to have the shape (batch_size, sequence_length, hidden_size).

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
def construct(self, inp):
    """
    Constructs the XLNet feed-forward layer.

    Args:
        self (XLNetFeedForward): The instance of the XLNetFeedForward class.
        inp: The input tensor to the feed-forward layer.

    Returns:
        None

    Raises:
        None

    This method applies the XLNet feed-forward layer operations on the input tensor.
    It performs the following steps:

    1. Applies layer_1 on the input tensor.
    2. Applies the activation function on the output of layer_1.
    3. Applies dropout regularization on the output of the activation function.
    4. Applies layer_2 on the output of the dropout operation.
    5. Applies dropout regularization on the output of layer_2.
    6. Adds the input tensor to the output of layer_2 and applies layer normalization.
    7. Returns the final output tensor.

    Note:
        The input tensor is expected to have the shape (batch_size, sequence_length, hidden_size).
    """
    output = inp
    output = self.layer_1(output)
    output = self.activation_function(output)
    output = self.dropout(output)
    output = self.layer_2(output)
    output = self.dropout(output)
    output = self.layer_norm(output + inp)
    return output

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForMultipleChoice

Bases: XLNetPreTrainedModel

This class represents an XLNet model for multiple choice tasks. It extends the XLNetPreTrainedModel class and provides functionality for constructing the model and handling multiple choice classification tasks. The class includes methods for initializing the model with configuration, constructing the model with input tensors, and computing the loss for multiple choice classification. It utilizes XLNetModel and SequenceSummary modules for processing input data and generating model outputs. The class also incorporates various input and output options to customize the model behavior during training and evaluation.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
class XLNetForMultipleChoice(XLNetPreTrainedModel):

    """
    This class represents an XLNet model for multiple choice tasks. It extends the XLNetPreTrainedModel class and
    provides functionality for constructing the model and handling multiple choice classification tasks. The class
    includes methods for initializing the model with configuration, constructing the model with input tensors, and
    computing the loss for multiple choice classification. It utilizes XLNetModel and SequenceSummary modules for
    processing input data and generating model outputs. The class also incorporates various input and output options to
    customize the model behavior during training and evaluation.
    """
    def __init__(self, config):
        """
        Initialize the XLNetForMultipleChoice class.

        Args:
            self (XLNetForMultipleChoice): The instance of the XLNetForMultipleChoice class.
            config: The configuration object containing parameters for model initialization.
               This should be an instance of a configuration class compatible with XLNetModel.

        Returns:
            None.

        Raises:
            TypeError: If the provided config is not of the expected type.
            ValueError: If there are any issues during the initialization process.
        """
        super().__init__(config)

        self.transformer = XLNetModel(config)
        self.sequence_summary = SequenceSummary(config)
        self.logits_proj = nn.Dense(config.d_model, 1)

        # Initialize weights and apply final processing
        self.post_init()

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            labels: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetForMultipleChoiceOutput]:
        r"""
        Args:
            labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
                Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
                num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
                `input_ids` above)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]

        flat_input_ids = input_ids.view(-1, input_ids.shape[-1]) if input_ids is not None else None
        flat_token_type_ids = token_type_ids.view(-1, token_type_ids.shape[-1]) if token_type_ids is not None else None
        flat_attention_mask = attention_mask.view(-1, attention_mask.shape[-1]) if attention_mask is not None else None
        flat_input_mask = input_mask.view(-1, input_mask.shape[-1]) if input_mask is not None else None
        flat_inputs_embeds = (
            inputs_embeds.view(-1, inputs_embeds.shape[-2], inputs_embeds.shape[-1])
            if inputs_embeds is not None
            else None
        )

        transformer_outputs = self.transformer(
            flat_input_ids,
            token_type_ids=flat_token_type_ids,
            input_mask=flat_input_mask,
            attention_mask=flat_attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            head_mask=head_mask,
            inputs_embeds=flat_inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            **kwargs,
        )

        output = transformer_outputs[0]

        output = self.sequence_summary(output)
        logits = self.logits_proj(output)
        reshaped_logits = logits.view(-1, num_choices)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            labels = labels.astype(mindspore.int32)
            loss = loss_fct(reshaped_logits, labels.view(-1))

        if not return_dict:
            output = (reshaped_logits,) + transformer_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return XLNetForMultipleChoiceOutput(
            loss=loss,
            logits=reshaped_logits,
            mems=transformer_outputs.mems,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForMultipleChoice.__init__(config)

Initialize the XLNetForMultipleChoice class.

PARAMETER DESCRIPTION
self

The instance of the XLNetForMultipleChoice class.

TYPE: XLNetForMultipleChoice

config

The configuration object containing parameters for model initialization. This should be an instance of a configuration class compatible with XLNetModel.
RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
TypeError

If the provided config is not of the expected type.
ValueError

If there are any issues during the initialization process.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
def __init__(self, config):
    """
    Initialize the XLNetForMultipleChoice class.

    Args:
        self (XLNetForMultipleChoice): The instance of the XLNetForMultipleChoice class.
        config: The configuration object containing parameters for model initialization.
           This should be an instance of a configuration class compatible with XLNetModel.

    Returns:
        None.

    Raises:
        TypeError: If the provided config is not of the expected type.
        ValueError: If there are any issues during the initialization process.
    """
    super().__init__(config)

    self.transformer = XLNetModel(config)
    self.sequence_summary = SequenceSummary(config)
    self.logits_proj = nn.Dense(config.d_model, 1)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForMultipleChoice.construct(input_ids=None, token_type_ids=None, input_mask=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, head_mask=None, inputs_embeds=None, labels=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
labels

Labels for computing the multiple choice classification loss. Indices should be in [0, ..., num_choices-1] where num_choices is the size of the second dimension of the input tensors. (See input_ids above)

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        labels: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetForMultipleChoiceOutput]:
    r"""
    Args:
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
            `input_ids` above)
    """
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]

    flat_input_ids = input_ids.view(-1, input_ids.shape[-1]) if input_ids is not None else None
    flat_token_type_ids = token_type_ids.view(-1, token_type_ids.shape[-1]) if token_type_ids is not None else None
    flat_attention_mask = attention_mask.view(-1, attention_mask.shape[-1]) if attention_mask is not None else None
    flat_input_mask = input_mask.view(-1, input_mask.shape[-1]) if input_mask is not None else None
    flat_inputs_embeds = (
        inputs_embeds.view(-1, inputs_embeds.shape[-2], inputs_embeds.shape[-1])
        if inputs_embeds is not None
        else None
    )

    transformer_outputs = self.transformer(
        flat_input_ids,
        token_type_ids=flat_token_type_ids,
        input_mask=flat_input_mask,
        attention_mask=flat_attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        head_mask=head_mask,
        inputs_embeds=flat_inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        **kwargs,
    )

    output = transformer_outputs[0]

    output = self.sequence_summary(output)
    logits = self.logits_proj(output)
    reshaped_logits = logits.view(-1, num_choices)

    loss = None
    if labels is not None:
        loss_fct = CrossEntropyLoss()
        labels = labels.astype(mindspore.int32)
        loss = loss_fct(reshaped_logits, labels.view(-1))

    if not return_dict:
        output = (reshaped_logits,) + transformer_outputs[1:]
        return ((loss,) + output) if loss is not None else output

    return XLNetForMultipleChoiceOutput(
        loss=loss,
        logits=reshaped_logits,
        mems=transformer_outputs.mems,
        hidden_states=transformer_outputs.hidden_states,
        attentions=transformer_outputs.attentions,
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForMultipleChoiceOutput dataclass

Bases: ModelOutput

Output type of [XLNetForMultipleChoice].

PARAMETER DESCRIPTION
loss

Classification loss.

TYPE: `torch.FloatTensor` of shape *(1,)*, *optional*, returned when `labels` is provided DEFAULT: None

logits

num_choices is the second dimension of the input tensors. (see input_ids above).

Classification scores (before SoftMax).

TYPE: `torch.FloatTensor` of shape `(batch_size, num_choices)` DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
@dataclass
class XLNetForMultipleChoiceOutput(ModelOutput):
    """
    Output type of [`XLNetForMultipleChoice`].

    Args:
        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).
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    logits: mindspore.Tensor = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnswering

Bases: XLNetPreTrainedModel

The XLNetForQuestionAnswering class represents a XLNet model for question answering. It inherits from XLNetPreTrainedModel and provides methods for constructing the model and processing input data for question answering tasks. The class includes methods for computing start and end positions of the labelled span, determining if a question has an answer or no answer, and computing the plausibility of the answer. Additionally, it provides functionality for handling optional masks of tokens that can't be in answers.

Example 
>>> from transformers import AutoTokenizer, XLNetForQuestionAnswering
>>> import torch
...
>>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-base-cased")
>>> model = XLNetForQuestionAnswering.from_pretrained("xlnet/xlnet-base-cased")
...
>>> 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
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
class XLNetForQuestionAnswering(XLNetPreTrainedModel):

    """
    The XLNetForQuestionAnswering class represents a XLNet model for question answering.
    It inherits from XLNetPreTrainedModel and provides methods for constructing the model and processing input data for
    question answering tasks. The class includes methods for computing start and end positions of the labelled span,
    determining if a question has an answer or no answer, and computing the plausibility of the answer. Additionally,
    it provides functionality for handling optional masks of tokens that can't be in answers.

    Example:
        ```python
        >>> from transformers import AutoTokenizer, XLNetForQuestionAnswering
        >>> import torch
        ...
        >>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-base-cased")
        >>> model = XLNetForQuestionAnswering.from_pretrained("xlnet/xlnet-base-cased")
        ...
        >>> 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
        ```
    """
    def __init__(self, config):
        """
        Initializes an instance of XLNetForQuestionAnswering.

        Args:
            self (XLNetForQuestionAnswering): The instance of the XLNetForQuestionAnswering class.
            config:
                An object containing configuration settings for XLNetForQuestionAnswering.

                - Type: Any
                - Purpose: Specifies the configuration settings to initialize the XLNetForQuestionAnswering instance.
                - Restrictions: Must be a valid configuration object.

        Returns:
            None.

        Raises:
            None.
        """
        super().__init__(config)
        self.start_n_top = config.start_n_top
        self.end_n_top = config.end_n_top

        self.transformer = XLNetModel(config)
        self.start_logits = PoolerStartLogits(config)
        self.end_logits = PoolerEndLogits(config)
        self.answer_class = PoolerAnswerClass(config)

        # Initialize weights and apply final processing
        self.post_init()

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            start_positions: Optional[mindspore.Tensor] = None,
            end_positions: Optional[mindspore.Tensor] = None,
            is_impossible: Optional[mindspore.Tensor] = None,
            cls_index: Optional[mindspore.Tensor] = None,
            p_mask: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetForQuestionAnsweringOutput]:
        r"""
        Args:
            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:
            `Union[Tuple, XLNetForQuestionAnsweringOutput]`

        Example:
            ```python
            >>> from transformers import AutoTokenizer, XLNetForQuestionAnswering
            >>> import torch
            ...
            >>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-base-cased")
            >>> model = XLNetForQuestionAnswering.from_pretrained("xlnet/xlnet-base-cased")
            ...
            >>> 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_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            token_type_ids=token_type_ids,
            input_mask=input_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            **kwargs,
        )
        hidden_states = transformer_outputs[0]
        start_logits = self.start_logits(hidden_states, p_mask=p_mask)

        outputs = transformer_outputs[1:]  # Keep mems, hidden states, attentions if there are in it

        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, let's remove the dimension added by batch splitting
            for x in (start_positions, end_positions, cls_index, is_impossible):
                if x is not None and x.dim() > 1:
                    x.squeeze_(-1)

            # during training, compute the end logits based on the ground truth of the start position
            end_logits = self.end_logits(hidden_states, start_positions=start_positions, p_mask=p_mask)

            loss_fct = CrossEntropyLoss()
            start_positions = start_positions.astype(mindspore.int32)
            end_positions = end_positions.astype(mindspore.int32)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2

            if cls_index is not None and is_impossible is not None:
                # Predict answerability from the representation of CLS and START
                cls_logits = self.answer_class(hidden_states, start_positions=start_positions, cls_index=cls_index)
                loss_fct_cls = nn.BCEWithLogitsLoss()
                cls_loss = loss_fct_cls(cls_logits, is_impossible)

                # note(zhiliny): by default multiply the loss by 0.5 so that the scale is comparable to start_loss and end_loss
                total_loss += cls_loss * 0.5

            if not return_dict:
                return (total_loss,) + transformer_outputs[1:]
            else:
                return XLNetForQuestionAnsweringOutput(
                    loss=total_loss,
                    mems=transformer_outputs.mems,
                    hidden_states=transformer_outputs.hidden_states,
                    attentions=transformer_outputs.attentions,
                )

        else:
            # during inference, compute the end logits based on beam search
            bsz, slen, hsz = hidden_states.shape
            start_log_probs = ops.softmax(start_logits, axis=-1)  # shape (bsz, slen)

            start_top_log_probs, start_top_index = ops.topk(
                start_log_probs, self.start_n_top, dim=-1
            )  # shape (bsz, start_n_top)
            start_top_index_exp = start_top_index.unsqueeze(-1).expand(-1, -1, hsz)  # shape (bsz, start_n_top, hsz)
            start_states = ops.gather_elements(hidden_states, -2, start_top_index_exp)  # shape (bsz, start_n_top, hsz)
            start_states = start_states.unsqueeze(1).expand(-1, slen, -1, -1)  # shape (bsz, slen, start_n_top, hsz)

            hidden_states_expanded = hidden_states.unsqueeze(2).expand_as(
                start_states
            )  # shape (bsz, slen, start_n_top, hsz)
            p_mask = p_mask.unsqueeze(-1) if p_mask is not None else None
            end_logits = self.end_logits(hidden_states_expanded, start_states=start_states, p_mask=p_mask)
            end_log_probs = ops.softmax(end_logits, axis=1)  # shape (bsz, slen, start_n_top)

            end_top_log_probs, end_top_index = ops.topk(
                end_log_probs, self.end_n_top, dim=1
            )  # shape (bsz, end_n_top, start_n_top)
            end_top_log_probs = end_top_log_probs.view(-1, self.start_n_top * self.end_n_top)
            end_top_index = end_top_index.view(-1, self.start_n_top * self.end_n_top)

            start_states = ops.einsum(
                "blh,bl->bh", hidden_states, start_log_probs
            )  # get the representation of START as weighted sum of hidden states
            cls_logits = self.answer_class(
                hidden_states, start_states=start_states, cls_index=cls_index
            )  # Shape (batch size,): one single `cls_logits` for each sample

            if not return_dict:
                outputs = (start_top_log_probs, start_top_index, end_top_log_probs, end_top_index, cls_logits)
                return outputs + transformer_outputs[1:]
            else:
                return XLNetForQuestionAnsweringOutput(
                    start_top_log_probs=start_top_log_probs,
                    start_top_index=start_top_index,
                    end_top_log_probs=end_top_log_probs,
                    end_top_index=end_top_index,
                    cls_logits=cls_logits,
                    mems=transformer_outputs.mems,
                    hidden_states=transformer_outputs.hidden_states,
                    attentions=transformer_outputs.attentions,
                )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnswering.__init__(config)

Initializes an instance of XLNetForQuestionAnswering.

PARAMETER DESCRIPTION
self

The instance of the XLNetForQuestionAnswering class.

TYPE: XLNetForQuestionAnswering

config

An object containing configuration settings for XLNetForQuestionAnswering.

  • Type: Any
  • Purpose: Specifies the configuration settings to initialize the XLNetForQuestionAnswering instance.
  • Restrictions: Must be a valid configuration object.
RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
def __init__(self, config):
    """
    Initializes an instance of XLNetForQuestionAnswering.

    Args:
        self (XLNetForQuestionAnswering): The instance of the XLNetForQuestionAnswering class.
        config:
            An object containing configuration settings for XLNetForQuestionAnswering.

            - Type: Any
            - Purpose: Specifies the configuration settings to initialize the XLNetForQuestionAnswering instance.
            - Restrictions: Must be a valid configuration object.

    Returns:
        None.

    Raises:
        None.
    """
    super().__init__(config)
    self.start_n_top = config.start_n_top
    self.end_n_top = config.end_n_top

    self.transformer = XLNetModel(config)
    self.start_logits = PoolerStartLogits(config)
    self.end_logits = PoolerEndLogits(config)
    self.answer_class = PoolerAnswerClass(config)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnswering.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, start_positions=None, end_positions=None, is_impossible=None, cls_index=None, p_mask=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
start_positions

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.

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

end_positions

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.

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

is_impossible

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

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

cls_index

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

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

p_mask

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.

TYPE: `torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional* DEFAULT: None

RETURNS DESCRIPTION
Union[Tuple, XLNetForQuestionAnsweringOutput]

Union[Tuple, XLNetForQuestionAnsweringOutput]
Example 
>>> from transformers import AutoTokenizer, XLNetForQuestionAnswering
>>> import torch
...
>>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-base-cased")
>>> model = XLNetForQuestionAnswering.from_pretrained("xlnet/xlnet-base-cased")
...
>>> 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
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        start_positions: Optional[mindspore.Tensor] = None,
        end_positions: Optional[mindspore.Tensor] = None,
        is_impossible: Optional[mindspore.Tensor] = None,
        cls_index: Optional[mindspore.Tensor] = None,
        p_mask: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetForQuestionAnsweringOutput]:
    r"""
    Args:
        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:
        `Union[Tuple, XLNetForQuestionAnsweringOutput]`

    Example:
        ```python
        >>> from transformers import AutoTokenizer, XLNetForQuestionAnswering
        >>> import torch
        ...
        >>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-base-cased")
        >>> model = XLNetForQuestionAnswering.from_pretrained("xlnet/xlnet-base-cased")
        ...
        >>> 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_dict = return_dict if return_dict is not None else self.config.use_return_dict

    transformer_outputs = self.transformer(
        input_ids,
        attention_mask=attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        token_type_ids=token_type_ids,
        input_mask=input_mask,
        head_mask=head_mask,
        inputs_embeds=inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        **kwargs,
    )
    hidden_states = transformer_outputs[0]
    start_logits = self.start_logits(hidden_states, p_mask=p_mask)

    outputs = transformer_outputs[1:]  # Keep mems, hidden states, attentions if there are in it

    if start_positions is not None and end_positions is not None:
        # If we are on multi-GPU, let's remove the dimension added by batch splitting
        for x in (start_positions, end_positions, cls_index, is_impossible):
            if x is not None and x.dim() > 1:
                x.squeeze_(-1)

        # during training, compute the end logits based on the ground truth of the start position
        end_logits = self.end_logits(hidden_states, start_positions=start_positions, p_mask=p_mask)

        loss_fct = CrossEntropyLoss()
        start_positions = start_positions.astype(mindspore.int32)
        end_positions = end_positions.astype(mindspore.int32)
        start_loss = loss_fct(start_logits, start_positions)
        end_loss = loss_fct(end_logits, end_positions)
        total_loss = (start_loss + end_loss) / 2

        if cls_index is not None and is_impossible is not None:
            # Predict answerability from the representation of CLS and START
            cls_logits = self.answer_class(hidden_states, start_positions=start_positions, cls_index=cls_index)
            loss_fct_cls = nn.BCEWithLogitsLoss()
            cls_loss = loss_fct_cls(cls_logits, is_impossible)

            # note(zhiliny): by default multiply the loss by 0.5 so that the scale is comparable to start_loss and end_loss
            total_loss += cls_loss * 0.5

        if not return_dict:
            return (total_loss,) + transformer_outputs[1:]
        else:
            return XLNetForQuestionAnsweringOutput(
                loss=total_loss,
                mems=transformer_outputs.mems,
                hidden_states=transformer_outputs.hidden_states,
                attentions=transformer_outputs.attentions,
            )

    else:
        # during inference, compute the end logits based on beam search
        bsz, slen, hsz = hidden_states.shape
        start_log_probs = ops.softmax(start_logits, axis=-1)  # shape (bsz, slen)

        start_top_log_probs, start_top_index = ops.topk(
            start_log_probs, self.start_n_top, dim=-1
        )  # shape (bsz, start_n_top)
        start_top_index_exp = start_top_index.unsqueeze(-1).expand(-1, -1, hsz)  # shape (bsz, start_n_top, hsz)
        start_states = ops.gather_elements(hidden_states, -2, start_top_index_exp)  # shape (bsz, start_n_top, hsz)
        start_states = start_states.unsqueeze(1).expand(-1, slen, -1, -1)  # shape (bsz, slen, start_n_top, hsz)

        hidden_states_expanded = hidden_states.unsqueeze(2).expand_as(
            start_states
        )  # shape (bsz, slen, start_n_top, hsz)
        p_mask = p_mask.unsqueeze(-1) if p_mask is not None else None
        end_logits = self.end_logits(hidden_states_expanded, start_states=start_states, p_mask=p_mask)
        end_log_probs = ops.softmax(end_logits, axis=1)  # shape (bsz, slen, start_n_top)

        end_top_log_probs, end_top_index = ops.topk(
            end_log_probs, self.end_n_top, dim=1
        )  # shape (bsz, end_n_top, start_n_top)
        end_top_log_probs = end_top_log_probs.view(-1, self.start_n_top * self.end_n_top)
        end_top_index = end_top_index.view(-1, self.start_n_top * self.end_n_top)

        start_states = ops.einsum(
            "blh,bl->bh", hidden_states, start_log_probs
        )  # get the representation of START as weighted sum of hidden states
        cls_logits = self.answer_class(
            hidden_states, start_states=start_states, cls_index=cls_index
        )  # Shape (batch size,): one single `cls_logits` for each sample

        if not return_dict:
            outputs = (start_top_log_probs, start_top_index, end_top_log_probs, end_top_index, cls_logits)
            return outputs + transformer_outputs[1:]
        else:
            return XLNetForQuestionAnsweringOutput(
                start_top_log_probs=start_top_log_probs,
                start_top_index=start_top_index,
                end_top_log_probs=end_top_log_probs,
                end_top_index=end_top_index,
                cls_logits=cls_logits,
                mems=transformer_outputs.mems,
                hidden_states=transformer_outputs.hidden_states,
                attentions=transformer_outputs.attentions,
            )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnsweringOutput dataclass

Bases: ModelOutput

Output type of [XLNetForQuestionAnswering].

PARAMETER DESCRIPTION
loss

Classification loss as the sum of start token, end token (and is_impossible if provided) classification losses.

TYPE: `torch.FloatTensor` of shape `(1,)`, *optional*, returned if both `start_positions` and `end_positions` are provided DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
@dataclass
class XLNetForQuestionAnsweringOutput(ModelOutput):
    """
    Output type of [`XLNetForQuestionAnswering`].

    Args:
        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.
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    start_top_log_probs: Optional[mindspore.Tensor] = None
    start_top_index: Optional[mindspore.Tensor] = None
    end_top_log_probs: Optional[mindspore.Tensor] = None
    end_top_index: Optional[mindspore.Tensor] = None
    cls_logits: Optional[mindspore.Tensor] = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnsweringSimple

Bases: XLNetPreTrainedModel

This class represents a simple implementation of the XLNet model for question answering tasks. It is designed specifically for question answering tasks where the start and end positions of the answer in the input sequence need to be predicted.

The XLNetForQuestionAnsweringSimple class inherits from the XLNetPreTrainedModel class, which provides the basic infrastructure and functionality for XLNet models.

The class has a constructor method __init__ that initializes the XLNetForQuestionAnsweringSimple instance with the given configuration. The configuration includes the number of labels for the classification task and other model-specific settings. It also initializes the XLNetModel transformer, which is responsible for the main computations of the XLNet model, and the qa_outputs module, which is a fully connected layer for predicting start and end positions.

The construct method is the main entry point for using the XLNetForQuestionAnsweringSimple model. It takes various input tensors, such as input_ids, attention_mask, and token_type_ids, which represent the input sequence and its properties. It also takes optional tensors such as start_positions and end_positions, which are the labels for the positions of the start and end of the answer span in the input sequence.

The method returns either a tuple or a XLNetForQuestionAnsweringSimpleOutput object, depending on the return_dict parameter. The output contains the predicted start and end logits, and optionally, the total loss, the transformer's mems, hidden states, and attentions.

The construct method also handles the computation of the loss if the start and end positions are provided. It clamps the positions to the length of the sequence and applies the CrossEntropyLoss to calculate the start and end losses. The total loss is the average of the start and end losses.

If the return_dict parameter is False, the method returns a tuple containing the total loss (if available), the start logits, the end logits, and other optional outputs. If the total loss is not available, the tuple contains only the logits and optional outputs.

If the return_dict parameter is True, the method returns a XLNetForQuestionAnsweringSimpleOutput object that encapsulates all the outputs.

Note

The class assumes the usage of the mindspore library for tensor operations and loss computation.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
class XLNetForQuestionAnsweringSimple(XLNetPreTrainedModel):

    """
    This class represents a simple implementation of the XLNet model for question answering tasks.
    It is designed specifically for question answering tasks where the start and end positions of the answer in
    the input sequence need to be predicted.

    The `XLNetForQuestionAnsweringSimple` class inherits from the `XLNetPreTrainedModel` class, which provides the
    basic infrastructure and functionality for XLNet models.

    The class has a constructor method `__init__` that initializes the XLNetForQuestionAnsweringSimple instance with
    the given configuration. The configuration includes the number of labels for the classification task and other
    model-specific settings. It also initializes the XLNetModel transformer, which is responsible for the main
    computations of the XLNet model, and the `qa_outputs` module, which is a fully connected layer for predicting start
    and end positions.

    The `construct` method is the main entry point for using the XLNetForQuestionAnsweringSimple model. It takes various
    input tensors, such as `input_ids`, `attention_mask`, and `token_type_ids`, which represent the input sequence and
    its properties. It also takes optional tensors such as `start_positions` and `end_positions`, which are the labels
    for the positions of the start and end of the answer span in the input sequence.

    The method returns either a tuple or a `XLNetForQuestionAnsweringSimpleOutput` object, depending on the `return_dict`
    parameter. The output contains the predicted start and end logits, and optionally, the total loss, the transformer's
    mems, hidden states, and attentions.

    The `construct` method also handles the computation of the loss if the start and end positions are provided.
    It clamps the positions to the length of the sequence and applies the CrossEntropyLoss to calculate the start and
    end losses. The total loss is the average of the start and end losses.

    If the `return_dict` parameter is `False`, the method returns a tuple containing the total loss (if available),
    the start logits, the end logits, and other optional outputs. If the total loss is not available, the tuple contains
    only the logits and optional outputs.

    If the `return_dict` parameter is `True`, the method returns a `XLNetForQuestionAnsweringSimpleOutput` object that
    encapsulates all the outputs.

    Note:
        The class assumes the usage of the `mindspore` library for tensor operations and loss computation.

    """
    def __init__(self, config):
        """
        Initializes an instance of the 'XLNetForQuestionAnsweringSimple' class.

        Args:
            self: The instance of the class.
            config (XLNetConfig): The configuration object for the XLNet model.
                The 'config' object should contain the following attributes:

                - num_labels (int): The number of labels for the classification task.
                This is used to initialize the 'qa_outputs' layer.
                - hidden_size (int): The size of the hidden state in the transformer model.
                This is used to initialize the 'qa_outputs' layer.

        Returns:
            None

        Raises:
            None
        """
        super().__init__(config)
        self.num_labels = config.num_labels

        self.transformer = XLNetModel(config)
        self.qa_outputs = nn.Dense(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            start_positions: Optional[mindspore.Tensor] = None,
            end_positions: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetForQuestionAnsweringSimpleOutput]:
        r"""
        Args:
            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.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            token_type_ids=token_type_ids,
            input_mask=input_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            **kwargs,
        )

        sequence_output = outputs[0]

        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, axis=-1)
        start_logits = start_logits.squeeze(-1)
        end_logits = end_logits.squeeze(-1)

        total_loss = None
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.shape) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.shape) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.shape[1]
            start_positions = start_positions.clamp(0, ignored_index)
            end_positions = end_positions.clamp(0, ignored_index)

            start_positions = start_positions.astype(mindspore.int32)
            end_positions = end_positions.astype(mindspore.int32)

            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2

        if not return_dict:
            output = (start_logits, end_logits) + outputs[1:]
            return ((total_loss,) + output) if total_loss is not None else output

        return XLNetForQuestionAnsweringSimpleOutput(
            loss=total_loss,
            start_logits=start_logits,
            end_logits=end_logits,
            mems=outputs.mems,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnsweringSimple.__init__(config)

Initializes an instance of the 'XLNetForQuestionAnsweringSimple' class.

PARAMETER DESCRIPTION
self

The instance of the class.
config

The configuration object for the XLNet model. The 'config' object should contain the following attributes:

  • num_labels (int): The number of labels for the classification task. This is used to initialize the 'qa_outputs' layer.
  • hidden_size (int): The size of the hidden state in the transformer model. This is used to initialize the 'qa_outputs' layer.

TYPE: XLNetConfig

RETURNS DESCRIPTION

None
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
def __init__(self, config):
    """
    Initializes an instance of the 'XLNetForQuestionAnsweringSimple' class.

    Args:
        self: The instance of the class.
        config (XLNetConfig): The configuration object for the XLNet model.
            The 'config' object should contain the following attributes:

            - num_labels (int): The number of labels for the classification task.
            This is used to initialize the 'qa_outputs' layer.
            - hidden_size (int): The size of the hidden state in the transformer model.
            This is used to initialize the 'qa_outputs' layer.

    Returns:
        None

    Raises:
        None
    """
    super().__init__(config)
    self.num_labels = config.num_labels

    self.transformer = XLNetModel(config)
    self.qa_outputs = nn.Dense(config.hidden_size, config.num_labels)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnsweringSimple.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, start_positions=None, end_positions=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
start_positions

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.

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

end_positions

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.

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        start_positions: Optional[mindspore.Tensor] = None,
        end_positions: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetForQuestionAnsweringSimpleOutput]:
    r"""
    Args:
        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.
    """
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    outputs = self.transformer(
        input_ids,
        attention_mask=attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        token_type_ids=token_type_ids,
        input_mask=input_mask,
        head_mask=head_mask,
        inputs_embeds=inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        **kwargs,
    )

    sequence_output = outputs[0]

    logits = self.qa_outputs(sequence_output)
    start_logits, end_logits = logits.split(1, axis=-1)
    start_logits = start_logits.squeeze(-1)
    end_logits = end_logits.squeeze(-1)

    total_loss = None
    if start_positions is not None and end_positions is not None:
        # If we are on multi-GPU, split add a dimension
        if len(start_positions.shape) > 1:
            start_positions = start_positions.squeeze(-1)
        if len(end_positions.shape) > 1:
            end_positions = end_positions.squeeze(-1)
        # sometimes the start/end positions are outside our model inputs, we ignore these terms
        ignored_index = start_logits.shape[1]
        start_positions = start_positions.clamp(0, ignored_index)
        end_positions = end_positions.clamp(0, ignored_index)

        start_positions = start_positions.astype(mindspore.int32)
        end_positions = end_positions.astype(mindspore.int32)

        loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
        start_loss = loss_fct(start_logits, start_positions)
        end_loss = loss_fct(end_logits, end_positions)
        total_loss = (start_loss + end_loss) / 2

    if not return_dict:
        output = (start_logits, end_logits) + outputs[1:]
        return ((total_loss,) + output) if total_loss is not None else output

    return XLNetForQuestionAnsweringSimpleOutput(
        loss=total_loss,
        start_logits=start_logits,
        end_logits=end_logits,
        mems=outputs.mems,
        hidden_states=outputs.hidden_states,
        attentions=outputs.attentions,
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForQuestionAnsweringSimpleOutput dataclass

Bases: ModelOutput

Output type of [XLNetForQuestionAnsweringSimple].

PARAMETER DESCRIPTION
loss

Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.

TYPE: `torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided DEFAULT: None

start_logits

Span-start scores (before SoftMax).

TYPE: `torch.FloatTensor` of shape `(batch_size, sequence_length,)` DEFAULT: None

end_logits

Span-end scores (before SoftMax).

TYPE: `torch.FloatTensor` of shape `(batch_size, sequence_length,)` DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
@dataclass
class XLNetForQuestionAnsweringSimpleOutput(ModelOutput):
    """
    Output type of [`XLNetForQuestionAnsweringSimple`].

    Args:
        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).
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    start_logits: mindspore.Tensor = None
    end_logits: mindspore.Tensor = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForSequenceClassification

Bases: XLNetPreTrainedModel

The XLNetForSequenceClassification class is a subclass of XLNetPreTrainedModel that represents a model for sequence classification tasks using XLNet.

XLNetForSequenceClassification utilizes the XLNet model architecture combined with a linear layer for classification. It can be used for both single-label and multi-label classification tasks.

To instantiate this class, you need to provide a config object as an argument. The config object contains various configuration parameters for the XLNet model and the classification layer.

METHOD DESCRIPTION
`construct`

This method constructs the XLNetForSequenceClassification model by performing the necessary computations. It takes several input tensors, such as input_ids, attention_mask, mems, perm_mask, target_mapping, token_type_ids, input_mask, head_mask, inputs_embeds, labels, and various optional arguments. It returns a tuple of outputs, including loss, logits, mems, hidden_states, and attentions.
ATTRIBUTE DESCRIPTION
`num_labels`

The number of labels in the classification task.
`config`

The configuration object containing parameters for the XLNet model and the classification layer.
`transformer`

The XLNetModel instance used for sequence representation.
`sequence_summary`

The SequenceSummary instance used to summarize the sequence representation.
`logits_proj`

The linear layer used to project the sequence summary to the number of labels.
Note
  • The construct method automatically determines the problem_type based on the config parameters and the provided labels. The problem_type can be either 'regression', 'single_label_classification', or 'multi_label_classification'.
  • The loss function used for regression is Mean-Square Loss (MSELoss), while for classification, it is Cross-Entropy Loss (CrossEntropyLoss) for single-label classification and Binary Cross-Entropy Loss (BCEWithLogitsLoss) for multi-label classification.
  • The construct method allows for various optional arguments, such as output_attentions, output_hidden_states, and return_dict, which control the output format of the XLNet model.
  • The construct method returns either a tuple of outputs if return_dict is False, or an instance of XLNetForSequenceClassificationOutput if return_dict is True.
Example 
>>> config = XLNetConfig(...)
>>> model = XLNetForSequenceClassification(config)
>>> inputs = {
...     'input_ids': input_ids,
...     'attention_mask': attention_mask,
...     'labels': labels,
... }
>>> outputs = model.construct(**inputs)
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
class XLNetForSequenceClassification(XLNetPreTrainedModel):

    """
    The `XLNetForSequenceClassification` class is a subclass of `XLNetPreTrainedModel` that represents a model for
    sequence classification tasks using XLNet.

    XLNetForSequenceClassification utilizes the XLNet model architecture combined with a linear layer for classification.
    It can be used for both single-label and multi-label classification tasks.

    To instantiate this class, you need to provide a `config` object as an argument. The `config` object contains various
    configuration parameters for the XLNet model and the classification layer.

    Methods:
        `construct`: This method constructs the XLNetForSequenceClassification model by performing the necessary
            computations. It takes several input tensors, such as `input_ids`, `attention_mask`, `mems`, `perm_mask`,
            `target_mapping`, `token_type_ids`, `input_mask`, `head_mask`, `inputs_embeds`, `labels`, and various
            optional arguments. It returns a tuple of outputs, including `loss`, `logits`, `mems`, `hidden_states`,
            and `attentions`.

    Attributes:
        `num_labels`: The number of labels in the classification task.
        `config`: The configuration object containing parameters for the XLNet model and the classification layer.
        `transformer`: The XLNetModel instance used for sequence representation.
        `sequence_summary`: The SequenceSummary instance used to summarize the sequence representation.
        `logits_proj`: The linear layer used to project the sequence summary to the number of labels.

    Note:
        - The `construct` method automatically determines the `problem_type` based on the `config` parameters and
        the provided `labels`. The `problem_type` can be either 'regression', 'single_label_classification',
        or 'multi_label_classification'.
        - The loss function used for regression is Mean-Square Loss (MSELoss), while for classification, it is
        Cross-Entropy Loss (CrossEntropyLoss) for single-label classification and Binary Cross-Entropy Loss
        (BCEWithLogitsLoss) for multi-label classification.
        - The `construct` method allows for various optional arguments, such as `output_attentions`,
        `output_hidden_states`, and `return_dict`, which control the output format of the XLNet model.
        - The `construct` method returns either a tuple of outputs if `return_dict` is False, or an instance of
        `XLNetForSequenceClassificationOutput` if `return_dict` is True.

    Example:
        ```python
        >>> config = XLNetConfig(...)
        >>> model = XLNetForSequenceClassification(config)
        >>> inputs = {
        ...     'input_ids': input_ids,
        ...     'attention_mask': attention_mask,
        ...     'labels': labels,
        ... }
        >>> outputs = model.construct(**inputs)
        ```
    """
    def __init__(self, config):
        """
        Initializes a new instance of the XLNetForSequenceClassification class.

        Args:
            self: The instance of the XLNetForSequenceClassification class.
            config: An instance of the XLNetConfig class containing the configuration settings for the XLNet model.

        Returns:
            None

        Raises:
            None
        """
        super().__init__(config)
        self.num_labels = config.num_labels
        self.config = config
        self.transformer = XLNetModel(config)
        self.sequence_summary = SequenceSummary(config)
        self.logits_proj = nn.Dense(config.d_model, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            labels: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetForSequenceClassificationOutput]:
        r"""
        Args:
            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).
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            token_type_ids=token_type_ids,
            input_mask=input_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            **kwargs,
        )
        output = transformer_outputs[0]

        output = self.sequence_summary(output)
        logits = self.logits_proj(output)

        loss = None
        if labels is not None:
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    self.config.problem_type = "regression"
                elif self.num_labels > 1 and (labels.dtype in (mindspore.int64, mindspore.int32)):
                    self.config.problem_type = "single_label_classification"
                else:
                    self.config.problem_type = "multi_label_classification"

            if self.config.problem_type == "regression":
                loss_fct = MSELoss()
                if self.num_labels == 1:
                    loss = loss_fct(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = CrossEntropyLoss()
                labels = labels.astype(mindspore.int32)
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)

        if not return_dict:
            output = (logits,) + transformer_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return XLNetForSequenceClassificationOutput(
            loss=loss,
            logits=logits,
            mems=transformer_outputs.mems,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForSequenceClassification.__init__(config)

Initializes a new instance of the XLNetForSequenceClassification class.

PARAMETER DESCRIPTION
self

The instance of the XLNetForSequenceClassification class.
config

An instance of the XLNetConfig class containing the configuration settings for the XLNet model.
RETURNS DESCRIPTION

None
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
def __init__(self, config):
    """
    Initializes a new instance of the XLNetForSequenceClassification class.

    Args:
        self: The instance of the XLNetForSequenceClassification class.
        config: An instance of the XLNetConfig class containing the configuration settings for the XLNet model.

    Returns:
        None

    Raises:
        None
    """
    super().__init__(config)
    self.num_labels = config.num_labels
    self.config = config
    self.transformer = XLNetModel(config)
    self.sequence_summary = SequenceSummary(config)
    self.logits_proj = nn.Dense(config.d_model, config.num_labels)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForSequenceClassification.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, labels=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
labels

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).

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        labels: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetForSequenceClassificationOutput]:
    r"""
    Args:
        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).
    """
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    transformer_outputs = self.transformer(
        input_ids,
        attention_mask=attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        token_type_ids=token_type_ids,
        input_mask=input_mask,
        head_mask=head_mask,
        inputs_embeds=inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        **kwargs,
    )
    output = transformer_outputs[0]

    output = self.sequence_summary(output)
    logits = self.logits_proj(output)

    loss = None
    if labels is not None:
        if self.config.problem_type is None:
            if self.num_labels == 1:
                self.config.problem_type = "regression"
            elif self.num_labels > 1 and (labels.dtype in (mindspore.int64, mindspore.int32)):
                self.config.problem_type = "single_label_classification"
            else:
                self.config.problem_type = "multi_label_classification"

        if self.config.problem_type == "regression":
            loss_fct = MSELoss()
            if self.num_labels == 1:
                loss = loss_fct(logits.squeeze(), labels.squeeze())
            else:
                loss = loss_fct(logits, labels)
        elif self.config.problem_type == "single_label_classification":
            loss_fct = CrossEntropyLoss()
            labels = labels.astype(mindspore.int32)
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
        elif self.config.problem_type == "multi_label_classification":
            loss_fct = BCEWithLogitsLoss()
            loss = loss_fct(logits, labels)

    if not return_dict:
        output = (logits,) + transformer_outputs[1:]
        return ((loss,) + output) if loss is not None else output

    return XLNetForSequenceClassificationOutput(
        loss=loss,
        logits=logits,
        mems=transformer_outputs.mems,
        hidden_states=transformer_outputs.hidden_states,
        attentions=transformer_outputs.attentions,
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForSequenceClassificationOutput dataclass

Bases: ModelOutput

Output type of [XLNetForSequenceClassification].

PARAMETER DESCRIPTION
loss

Classification (or regression if config.num_labels==1) loss.

TYPE: `torch.FloatTensor` of shape `(1,)`, *optional*, returned when `label` is provided DEFAULT: None

logits

Classification (or regression if config.num_labels==1) scores (before SoftMax).

TYPE: `torch.FloatTensor` of shape `(batch_size, config.num_labels)` DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
@dataclass
class XLNetForSequenceClassificationOutput(ModelOutput):
    """
    Output type of [`XLNetForSequenceClassification`].

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `label` 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).
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    logits: mindspore.Tensor = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForTokenClassification

Bases: XLNetPreTrainedModel

XLNetForTokenClassification is a class that represents a XLNet model for token classification tasks, inheriting from XLNetPreTrainedModel. It includes methods for initializing the model with configuration parameters, constructing the model with various input tensors and optional parameters, and computing the token classification loss.

ATTRIBUTE DESCRIPTION
num_labels

The number of labels for token classification.

TYPE: int

transformer

The XLNet model for processing input tensors.

TYPE: XLNetModel

classifier

The classifier layer for token classification.

TYPE: Dense

METHOD DESCRIPTION
__init__

Initializes the XLNetForTokenClassification instance with the provided configuration.
construct

Constructs the XLNetForTokenClassification model using the input tensors and optional parameters, and computes the token classification loss.

Parameters:

  • input_ids (mindspore.Tensor): The input tensor representing token IDs.
  • attention_mask (mindspore.Tensor): The tensor indicating attention masks.
  • mems (mindspore.Tensor): The tensor for memory inputs.
  • perm_mask (mindspore.Tensor): The tensor for permutation masks.
  • target_mapping (mindspore.Tensor): The tensor for target mapping.
  • token_type_ids (mindspore.Tensor): The tensor for token type IDs.
  • input_mask (mindspore.Tensor): The tensor indicating input masks.
  • head_mask (mindspore.Tensor): The tensor for head masks.
  • inputs_embeds (mindspore.Tensor): The tensor for input embeddings.
  • labels (mindspore.Tensor): The tensor for target labels.
  • use_mems (bool): Flag indicating whether to use memory inputs.
  • output_attentions (bool): Flag indicating whether to output attentions.
  • output_hidden_states (bool): Flag indicating whether to output hidden states.
  • return_dict (bool): Flag indicating whether to return output as a dictionary.
  • kwargs (dict): Additional keyword arguments.

Returns:

  • Union[Tuple, XLNetForTokenClassificationOutput]: A tuple or XLNetForTokenClassificationOutput object containing the computed loss, logits, memories, hidden states, and attentions.

Notes:

  • labels should be a torch.LongTensor with indices in [0, ..., num_choices].
  • Loss is computed using CrossEntropyLoss.

Example: 

>>> model = XLNetForTokenClassification(config)
>>> outputs = model.construct(input_ids=input_tensor, attention_mask=attention_mask, labels=label_tensor)

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
class XLNetForTokenClassification(XLNetPreTrainedModel):

    """
    XLNetForTokenClassification is a class that represents a XLNet model for token classification tasks, inheriting
    from XLNetPreTrainedModel.
    It includes methods for initializing the model with configuration parameters, constructing the model with various
    input tensors and optional parameters, and computing the token classification loss.

    Attributes:
        num_labels (int): The number of labels for token classification.
        transformer (XLNetModel): The XLNet model for processing input tensors.
        classifier (nn.Dense): The classifier layer for token classification.

    Methods:
        __init__: Initializes the XLNetForTokenClassification instance with the provided configuration.
        construct:
            Constructs the XLNetForTokenClassification model using the input tensors and optional parameters,
            and computes the token classification loss.

            Parameters:

            - input_ids (mindspore.Tensor): The input tensor representing token IDs.
            - attention_mask (mindspore.Tensor): The tensor indicating attention masks.
            - mems (mindspore.Tensor): The tensor for memory inputs.
            - perm_mask (mindspore.Tensor): The tensor for permutation masks.
            - target_mapping (mindspore.Tensor): The tensor for target mapping.
            - token_type_ids (mindspore.Tensor): The tensor for token type IDs.
            - input_mask (mindspore.Tensor): The tensor indicating input masks.
            - head_mask (mindspore.Tensor): The tensor for head masks.
            - inputs_embeds (mindspore.Tensor): The tensor for input embeddings.
            - labels (mindspore.Tensor): The tensor for target labels.
            - use_mems (bool): Flag indicating whether to use memory inputs.
            - output_attentions (bool): Flag indicating whether to output attentions.
            - output_hidden_states (bool): Flag indicating whether to output hidden states.
            - return_dict (bool): Flag indicating whether to return output as a dictionary.
            - kwargs (dict): Additional keyword arguments.

            Returns:

            - Union[Tuple, XLNetForTokenClassificationOutput]: A tuple or XLNetForTokenClassificationOutput object
            containing the computed loss, logits, memories, hidden states, and attentions.

            Notes:

            - labels should be a torch.LongTensor with indices in [0, ..., num_choices].
            - Loss is computed using CrossEntropyLoss.

            Example:
                ```python
                >>> model = XLNetForTokenClassification(config)
                >>> outputs = model.construct(input_ids=input_tensor, attention_mask=attention_mask, labels=label_tensor)
                ```
    """
    def __init__(self, config):
        """
        Initializes an instance of XLNetForTokenClassification.

        Args:
            self (XLNetForTokenClassification): The instance of the XLNetForTokenClassification class.
            config:
                A configuration object containing parameters for the model.

                - Type: Any
                - Purpose: Specifies the configuration settings for the model.
                - Restrictions: Must be compatible with the XLNetModel and nn.Dense classes.

        Returns:
            None.

        Raises:
            None
        """
        super().__init__(config)
        self.num_labels = config.num_labels

        self.transformer = XLNetModel(config)
        self.classifier = nn.Dense(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            labels: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetForTokenClassificationOutput]:
        r"""
        Args:
            labels (`torch.LongTensor` 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)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            token_type_ids=token_type_ids,
            input_mask=input_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            labels = labels.astype(mindspore.int32)
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return XLNetForTokenClassificationOutput(
            loss=loss,
            logits=logits,
            mems=outputs.mems,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForTokenClassification.__init__(config)

Initializes an instance of XLNetForTokenClassification.

PARAMETER DESCRIPTION
self

The instance of the XLNetForTokenClassification class.

TYPE: XLNetForTokenClassification

config

A configuration object containing parameters for the model.

  • Type: Any
  • Purpose: Specifies the configuration settings for the model.
  • Restrictions: Must be compatible with the XLNetModel and nn.Dense classes.
RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
def __init__(self, config):
    """
    Initializes an instance of XLNetForTokenClassification.

    Args:
        self (XLNetForTokenClassification): The instance of the XLNetForTokenClassification class.
        config:
            A configuration object containing parameters for the model.

            - Type: Any
            - Purpose: Specifies the configuration settings for the model.
            - Restrictions: Must be compatible with the XLNetModel and nn.Dense classes.

    Returns:
        None.

    Raises:
        None
    """
    super().__init__(config)
    self.num_labels = config.num_labels

    self.transformer = XLNetModel(config)
    self.classifier = nn.Dense(config.hidden_size, config.num_labels)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForTokenClassification.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, labels=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
labels

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)

TYPE: `torch.LongTensor` of shape `(batch_size,)`, *optional* DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        labels: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetForTokenClassificationOutput]:
    r"""
    Args:
        labels (`torch.LongTensor` 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)
    """
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    outputs = self.transformer(
        input_ids,
        attention_mask=attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        token_type_ids=token_type_ids,
        input_mask=input_mask,
        head_mask=head_mask,
        inputs_embeds=inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
    )

    sequence_output = outputs[0]

    logits = self.classifier(sequence_output)

    loss = None
    if labels is not None:
        loss_fct = CrossEntropyLoss()
        labels = labels.astype(mindspore.int32)
        loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

    if not return_dict:
        output = (logits,) + outputs[1:]
        return ((loss,) + output) if loss is not None else output

    return XLNetForTokenClassificationOutput(
        loss=loss,
        logits=logits,
        mems=outputs.mems,
        hidden_states=outputs.hidden_states,
        attentions=outputs.attentions,
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetForTokenClassificationOutput dataclass

Bases: ModelOutput

Output type of [XLNetForTokenClassificationOutput].

PARAMETER DESCRIPTION
loss

Classification loss.

TYPE: `torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided) DEFAULT: None

logits

Classification scores (before SoftMax).

TYPE: `torch.FloatTensor` of shape `(batch_size, sequence_length, config.num_labels)` DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
@dataclass
class XLNetForTokenClassificationOutput(ModelOutput):
    """
    Output type of [`XLNetForTokenClassificationOutput`].

    Args:
        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).
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    logits: mindspore.Tensor = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel

Bases: XLNetPreTrainedModel

A Python class representing the XLNetLMHeadModel, which inherits from XLNetPreTrainedModel.

XLNetLMHeadModel includes methods for initializing the model, preparing inputs for generation, and constructing the model for language modeling tasks. It also provides a method for reordering the cache during beam search or beam sample generation.

The XLNetLMHeadModel class is designed to work with XLNetModel and nn.Dense to process input data, generate predictions, and calculate loss during training.

The class includes methods for preparing inputs for language generation tasks, such as masked language modeling, and for constructing the model to perform auto-regressive language modeling.

The _reorder_cache method is used to re-order the mems cache during beam search or beam sample generation to match mems with the correct beam_idx at each generation step.

The class is designed to be used in conjunction with the XLNetModel and XLNetLMHeadModelOutput classes to facilitate language modeling tasks.

For usage examples and additional information, refer to the provided code documentation.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
class XLNetLMHeadModel(XLNetPreTrainedModel):

    """
    A Python class representing the XLNetLMHeadModel, which inherits from XLNetPreTrainedModel.

    XLNetLMHeadModel includes methods for initializing the model, preparing inputs for generation, and constructing
    the model for language modeling tasks. It also provides a method for reordering the cache during beam search or
    beam sample generation.

    The XLNetLMHeadModel class is designed to work with XLNetModel and nn.Dense to process input data, generate
    predictions, and calculate loss during training.

    The class includes methods for preparing inputs for language generation tasks, such as masked language modeling,
    and for constructing the model to perform auto-regressive language modeling.

    The _reorder_cache method is used to re-order the mems cache during beam search or beam sample generation to match
    mems with the correct beam_idx at each generation step.

    The class is designed to be used in conjunction with the XLNetModel and XLNetLMHeadModelOutput classes to facilitate
    language modeling tasks.

    For usage examples and additional information, refer to the provided code documentation.
    """
    _tied_weights_keys = ["lm_loss.weight"]

    def __init__(self, config):
        """
        Initializes an instance of the XLNetLMHeadModel class.

        Args:
            self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
            config (XLNetConfig): The configuration object for the XLNet model.

        Returns:
            None.

        Raises:
            ValueError: If the configuration is invalid or missing required attributes.
            TypeError: If the provided config is not of type XLNetConfig.
        """
        super().__init__(config)
        self.attn_type = config.attn_type
        self.same_length = config.same_length

        self.transformer = XLNetModel(config)
        self.lm_loss = nn.Dense(config.d_model, config.vocab_size, has_bias=True)

        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        """
        Returns the output embeddings of the XLNet language model head.

        Args:
            self: An instance of the XLNetLMHeadModel class.

        Returns:
            lm_loss: This method returns the output embeddings of the XLNet language model head.
                The output embeddings are used in various downstream tasks such as text classification
                and named entity recognition.

        Raises:
            None.
        """
        return self.lm_loss

    def set_output_embeddings(self, new_embeddings):
        """
        This method sets the output embeddings for the XLNetLMHeadModel.

        Args:
            self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
            new_embeddings (tensor): The new output embeddings to be set for the model.
                It should be a tensor of the appropriate shape and type.

        Returns:
            None.

        Raises:
            TypeError: If the new_embeddings parameter is not of type tensor.
            ValueError: If the new_embeddings parameter does not meet the required shape or type constraints.
        """
        self.lm_loss = new_embeddings

    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, use_mems=None, **kwargs):
        '''
        Args:
            self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
            input_ids (Tensor): The input tensor containing tokenized input IDs.
            past_key_values (tuple, optional): A tuple of past key values from previous generation steps.
                Defaults to None.
            use_mems (bool, optional): A boolean flag indicating whether to use memory. Defaults to None.

        Returns:
            None: This method does not return any value.

        Raises:
            ValueError: If the input_ids tensor is not of the expected shape.
            ValueError: If past_key_values are provided but are not in the expected format.
            TypeError: If use_mems is not a boolean value.
        '''
        # Add dummy token at the end (no attention on this one)

        effective_batch_size = input_ids.shape[0]
        dummy_token = ops.zeros((effective_batch_size, 1), dtype=mindspore.int64)

        # At every pass, the attention values for the new token and the two last generated tokens
        # are computed, the rest is reloaded from the `past` cache. A purely auto-regressive model would have
        # offset = 1; offset = 2 seems to have slightly better computation.
        offset = 2

        if past_key_values:
            input_ids = ops.cat([input_ids[:, -offset:], dummy_token], axis=1)
        else:
            input_ids = ops.cat([input_ids, dummy_token], axis=1)

        # Build permutation mask so that previous tokens don't see last token
        sequence_length = input_ids.shape[1]
        perm_mask = ops.zeros(
            (effective_batch_size, sequence_length, sequence_length), dtype=mindspore.float32
        )
        perm_mask[:, :, -1] = 1.0

        # We'll only predict the last token
        target_mapping = ops.zeros(
            (effective_batch_size, 1, sequence_length), dtype=mindspore.float32
        )
        target_mapping[:, 0, -1] = 1.0

        inputs = {
            "input_ids": input_ids,
            "perm_mask": perm_mask,
            "target_mapping": target_mapping,
            "use_mems": use_mems,
        }

        # if past is defined in model kwargs then use it for faster decoding
        if past_key_values:
            inputs["mems"] = tuple(layer_past[:-offset, :, :] for layer_past in past_key_values)

        return inputs

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            labels: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete when `use_cache` is removed in XLNetModel
    ) -> Union[Tuple, XLNetLMHeadModelOutput]:
        r"""
        Args:
            labels (`torch.LongTensor` of shape `(batch_size, num_predict)`, *optional*):
                Labels for masked language modeling. `num_predict` corresponds to `target_mapping.shape[1]`. If
                `target_mapping` is `None`, then `num_predict` corresponds to `sequence_length`.

                The labels should correspond to the masked input words that should be predicted and depends on
                `target_mapping`. Note in order to perform standard auto-regressive language modeling a *<mask>* token has
                to be added to the `input_ids` (see the `prepare_inputs_for_generation` function and examples below)

                Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100` are ignored, the loss
                is only computed for labels in `[0, ..., config.vocab_size]`

        Returns:
            `Union[Tuple, XLNetLMHeadModelOutput]`

        Example:
            ```python
            >>> from transformers import AutoTokenizer, XLNetLMHeadModel
            >>> import torch
            ...
            >>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-large-cased")
            >>> model = XLNetLMHeadModel.from_pretrained("xlnet/xlnet-large-cased")
            ...
            >>> # We show how to setup inputs to predict a next token using a bi-directional context.
            >>> input_ids = torch.tensor(
            ...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
            ... ).unsqueeze(
            ...     0
            ... )  # We will predict the masked token
            >>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
            >>> perm_mask[:, :, -1] = 1.0  # Previous tokens don't see last token
            >>> target_mapping = torch.zeros(
            ...     (1, 1, input_ids.shape[1]), dtype=torch.float
            ... )  # Shape [1, 1, seq_length] => let's predict one token
            >>> target_mapping[
            ...     0, 0, -1
            ... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
            ...
            >>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping)
            >>> next_token_logits = outputs[
            ...     0
            ... ]  # Output has shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
            ...
            >>> # The same way can the XLNetLMHeadModel be used to be trained by standard auto-regressive language modeling.
            >>> input_ids = torch.tensor(
            ...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
            ... ).unsqueeze(
            ...     0
            ... )  # We will predict the masked token
            >>> labels = torch.tensor(tokenizer.encode("cute", add_special_tokens=False)).unsqueeze(0)
            >>> assert labels.shape[0] == 1, "only one word will be predicted"
            >>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
            >>> perm_mask[
            ...     :, :, -1
            ... ] = 1.0  # Previous tokens don't see last token as is done in standard auto-regressive lm training
            >>> target_mapping = torch.zeros(
            ...     (1, 1, input_ids.shape[1]), dtype=torch.float
            ... )  # Shape [1, 1, seq_length] => let's predict one token
            >>> target_mapping[
            ...     0, 0, -1
            ... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
            ...
            >>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping, labels=labels)
            >>> loss = outputs.loss
            >>> next_token_logits = (
            ...     outputs.logits
            ... )  # Logits have shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
            ```
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            mems=mems,
            perm_mask=perm_mask,
            target_mapping=target_mapping,
            token_type_ids=token_type_ids,
            input_mask=input_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_mems=use_mems,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            **kwargs,
        )

        logits = self.lm_loss(transformer_outputs[0])

        loss = None
        if labels is not None:
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            labels = labels.astype(mindspore.int32)
            loss = loss_fct(logits.view(-1, logits.shape[-1]), labels.view(-1))

        if not return_dict:
            output = (logits,) + transformer_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return XLNetLMHeadModelOutput(
            loss=loss,
            logits=logits,
            mems=transformer_outputs.mems,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )

    @staticmethod
    def _reorder_cache(mems: List[mindspore.Tensor], beam_idx: mindspore.Tensor) -> List[mindspore.Tensor]:
        """
        This function is used to re-order the `mems` cache if [`~PreTrainedModel.beam_search`] or
        [`~PreTrainedModel.beam_sample`] is called. This is required to match `mems` with the correct beam_idx at every
        generation step.
        """
        return [layer_past.index_select(1, beam_idx) for layer_past in mems]

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel.__init__(config)

Initializes an instance of the XLNetLMHeadModel class.

PARAMETER DESCRIPTION
self

The instance of the XLNetLMHeadModel class.

TYPE: XLNetLMHeadModel

config

The configuration object for the XLNet model.

TYPE: XLNetConfig

RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
ValueError

If the configuration is invalid or missing required attributes.
TypeError

If the provided config is not of type XLNetConfig.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
def __init__(self, config):
    """
    Initializes an instance of the XLNetLMHeadModel class.

    Args:
        self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
        config (XLNetConfig): The configuration object for the XLNet model.

    Returns:
        None.

    Raises:
        ValueError: If the configuration is invalid or missing required attributes.
        TypeError: If the provided config is not of type XLNetConfig.
    """
    super().__init__(config)
    self.attn_type = config.attn_type
    self.same_length = config.same_length

    self.transformer = XLNetModel(config)
    self.lm_loss = nn.Dense(config.d_model, config.vocab_size, has_bias=True)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, labels=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
labels

Labels for masked language modeling. num_predict corresponds to target_mapping.shape[1]. If target_mapping is None, then num_predict corresponds to sequence_length.

The labels should correspond to the masked input words that should be predicted and depends on target_mapping. Note in order to perform standard auto-regressive language modeling a token has to be added to the input_ids (see the prepare_inputs_for_generation function and examples below)

Indices are selected in [-100, 0, ..., config.vocab_size] All labels set to -100 are ignored, the loss is only computed for labels in [0, ..., config.vocab_size]

TYPE: `torch.LongTensor` of shape `(batch_size, num_predict)`, *optional* DEFAULT: None

RETURNS DESCRIPTION
Union[Tuple, XLNetLMHeadModelOutput]

Union[Tuple, XLNetLMHeadModelOutput]
Example 
>>> from transformers import AutoTokenizer, XLNetLMHeadModel
>>> import torch
...
>>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-large-cased")
>>> model = XLNetLMHeadModel.from_pretrained("xlnet/xlnet-large-cased")
...
>>> # We show how to setup inputs to predict a next token using a bi-directional context.
>>> input_ids = torch.tensor(
...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
... ).unsqueeze(
...     0
... )  # We will predict the masked token
>>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
>>> perm_mask[:, :, -1] = 1.0  # Previous tokens don't see last token
>>> target_mapping = torch.zeros(
...     (1, 1, input_ids.shape[1]), dtype=torch.float
... )  # Shape [1, 1, seq_length] => let's predict one token
>>> target_mapping[
...     0, 0, -1
... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
...
>>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping)
>>> next_token_logits = outputs[
...     0
... ]  # Output has shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
...
>>> # The same way can the XLNetLMHeadModel be used to be trained by standard auto-regressive language modeling.
>>> input_ids = torch.tensor(
...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
... ).unsqueeze(
...     0
... )  # We will predict the masked token
>>> labels = torch.tensor(tokenizer.encode("cute", add_special_tokens=False)).unsqueeze(0)
>>> assert labels.shape[0] == 1, "only one word will be predicted"
>>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
>>> perm_mask[
...     :, :, -1
... ] = 1.0  # Previous tokens don't see last token as is done in standard auto-regressive lm training
>>> target_mapping = torch.zeros(
...     (1, 1, input_ids.shape[1]), dtype=torch.float
... )  # Shape [1, 1, seq_length] => let's predict one token
>>> target_mapping[
...     0, 0, -1
... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
...
>>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping, labels=labels)
>>> loss = outputs.loss
>>> next_token_logits = (
...     outputs.logits
... )  # Logits have shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        labels: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete when `use_cache` is removed in XLNetModel
) -> Union[Tuple, XLNetLMHeadModelOutput]:
    r"""
    Args:
        labels (`torch.LongTensor` of shape `(batch_size, num_predict)`, *optional*):
            Labels for masked language modeling. `num_predict` corresponds to `target_mapping.shape[1]`. If
            `target_mapping` is `None`, then `num_predict` corresponds to `sequence_length`.

            The labels should correspond to the masked input words that should be predicted and depends on
            `target_mapping`. Note in order to perform standard auto-regressive language modeling a *<mask>* token has
            to be added to the `input_ids` (see the `prepare_inputs_for_generation` function and examples below)

            Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100` are ignored, the loss
            is only computed for labels in `[0, ..., config.vocab_size]`

    Returns:
        `Union[Tuple, XLNetLMHeadModelOutput]`

    Example:
        ```python
        >>> from transformers import AutoTokenizer, XLNetLMHeadModel
        >>> import torch
        ...
        >>> tokenizer = AutoTokenizer.from_pretrained("xlnet/xlnet-large-cased")
        >>> model = XLNetLMHeadModel.from_pretrained("xlnet/xlnet-large-cased")
        ...
        >>> # We show how to setup inputs to predict a next token using a bi-directional context.
        >>> input_ids = torch.tensor(
        ...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
        ... ).unsqueeze(
        ...     0
        ... )  # We will predict the masked token
        >>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
        >>> perm_mask[:, :, -1] = 1.0  # Previous tokens don't see last token
        >>> target_mapping = torch.zeros(
        ...     (1, 1, input_ids.shape[1]), dtype=torch.float
        ... )  # Shape [1, 1, seq_length] => let's predict one token
        >>> target_mapping[
        ...     0, 0, -1
        ... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
        ...
        >>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping)
        >>> next_token_logits = outputs[
        ...     0
        ... ]  # Output has shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
        ...
        >>> # The same way can the XLNetLMHeadModel be used to be trained by standard auto-regressive language modeling.
        >>> input_ids = torch.tensor(
        ...     tokenizer.encode("Hello, my dog is very <mask>", add_special_tokens=False)
        ... ).unsqueeze(
        ...     0
        ... )  # We will predict the masked token
        >>> labels = torch.tensor(tokenizer.encode("cute", add_special_tokens=False)).unsqueeze(0)
        >>> assert labels.shape[0] == 1, "only one word will be predicted"
        >>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
        >>> perm_mask[
        ...     :, :, -1
        ... ] = 1.0  # Previous tokens don't see last token as is done in standard auto-regressive lm training
        >>> target_mapping = torch.zeros(
        ...     (1, 1, input_ids.shape[1]), dtype=torch.float
        ... )  # Shape [1, 1, seq_length] => let's predict one token
        >>> target_mapping[
        ...     0, 0, -1
        ... ] = 1.0  # Our first (and only) prediction will be the last token of the sequence (the masked token)
        ...
        >>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping, labels=labels)
        >>> loss = outputs.loss
        >>> next_token_logits = (
        ...     outputs.logits
        ... )  # Logits have shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
        ```
    """
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    transformer_outputs = self.transformer(
        input_ids,
        attention_mask=attention_mask,
        mems=mems,
        perm_mask=perm_mask,
        target_mapping=target_mapping,
        token_type_ids=token_type_ids,
        input_mask=input_mask,
        head_mask=head_mask,
        inputs_embeds=inputs_embeds,
        use_mems=use_mems,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        **kwargs,
    )

    logits = self.lm_loss(transformer_outputs[0])

    loss = None
    if labels is not None:
        # Flatten the tokens
        loss_fct = CrossEntropyLoss()
        labels = labels.astype(mindspore.int32)
        loss = loss_fct(logits.view(-1, logits.shape[-1]), labels.view(-1))

    if not return_dict:
        output = (logits,) + transformer_outputs[1:]
        return ((loss,) + output) if loss is not None else output

    return XLNetLMHeadModelOutput(
        loss=loss,
        logits=logits,
        mems=transformer_outputs.mems,
        hidden_states=transformer_outputs.hidden_states,
        attentions=transformer_outputs.attentions,
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel.get_output_embeddings()

Returns the output embeddings of the XLNet language model head.

PARAMETER DESCRIPTION
self

An instance of the XLNetLMHeadModel class.
RETURNS DESCRIPTION
lm_loss

This method returns the output embeddings of the XLNet language model head. The output embeddings are used in various downstream tasks such as text classification and named entity recognition.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
def get_output_embeddings(self):
    """
    Returns the output embeddings of the XLNet language model head.

    Args:
        self: An instance of the XLNetLMHeadModel class.

    Returns:
        lm_loss: This method returns the output embeddings of the XLNet language model head.
            The output embeddings are used in various downstream tasks such as text classification
            and named entity recognition.

    Raises:
        None.
    """
    return self.lm_loss

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel.prepare_inputs_for_generation(input_ids, past_key_values=None, use_mems=None, **kwargs)

PARAMETER DESCRIPTION
self

The instance of the XLNetLMHeadModel class.

TYPE: XLNetLMHeadModel

input_ids

The input tensor containing tokenized input IDs.

TYPE: Tensor

past_key_values

A tuple of past key values from previous generation steps. Defaults to None.

TYPE: tuple DEFAULT: None

use_mems

A boolean flag indicating whether to use memory. Defaults to None.

TYPE: bool DEFAULT: None

RETURNS DESCRIPTION
None

This method does not return any value.
RAISES DESCRIPTION
ValueError

If the input_ids tensor is not of the expected shape.
ValueError

If past_key_values are provided but are not in the expected format.
TypeError

If use_mems is not a boolean value.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
def prepare_inputs_for_generation(self, input_ids, past_key_values=None, use_mems=None, **kwargs):
    '''
    Args:
        self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
        input_ids (Tensor): The input tensor containing tokenized input IDs.
        past_key_values (tuple, optional): A tuple of past key values from previous generation steps.
            Defaults to None.
        use_mems (bool, optional): A boolean flag indicating whether to use memory. Defaults to None.

    Returns:
        None: This method does not return any value.

    Raises:
        ValueError: If the input_ids tensor is not of the expected shape.
        ValueError: If past_key_values are provided but are not in the expected format.
        TypeError: If use_mems is not a boolean value.
    '''
    # Add dummy token at the end (no attention on this one)

    effective_batch_size = input_ids.shape[0]
    dummy_token = ops.zeros((effective_batch_size, 1), dtype=mindspore.int64)

    # At every pass, the attention values for the new token and the two last generated tokens
    # are computed, the rest is reloaded from the `past` cache. A purely auto-regressive model would have
    # offset = 1; offset = 2 seems to have slightly better computation.
    offset = 2

    if past_key_values:
        input_ids = ops.cat([input_ids[:, -offset:], dummy_token], axis=1)
    else:
        input_ids = ops.cat([input_ids, dummy_token], axis=1)

    # Build permutation mask so that previous tokens don't see last token
    sequence_length = input_ids.shape[1]
    perm_mask = ops.zeros(
        (effective_batch_size, sequence_length, sequence_length), dtype=mindspore.float32
    )
    perm_mask[:, :, -1] = 1.0

    # We'll only predict the last token
    target_mapping = ops.zeros(
        (effective_batch_size, 1, sequence_length), dtype=mindspore.float32
    )
    target_mapping[:, 0, -1] = 1.0

    inputs = {
        "input_ids": input_ids,
        "perm_mask": perm_mask,
        "target_mapping": target_mapping,
        "use_mems": use_mems,
    }

    # if past is defined in model kwargs then use it for faster decoding
    if past_key_values:
        inputs["mems"] = tuple(layer_past[:-offset, :, :] for layer_past in past_key_values)

    return inputs

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModel.set_output_embeddings(new_embeddings)

This method sets the output embeddings for the XLNetLMHeadModel.

PARAMETER DESCRIPTION
self

The instance of the XLNetLMHeadModel class.

TYPE: XLNetLMHeadModel

new_embeddings

The new output embeddings to be set for the model. It should be a tensor of the appropriate shape and type.

TYPE: tensor

RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
TypeError

If the new_embeddings parameter is not of type tensor.
ValueError

If the new_embeddings parameter does not meet the required shape or type constraints.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
def set_output_embeddings(self, new_embeddings):
    """
    This method sets the output embeddings for the XLNetLMHeadModel.

    Args:
        self (XLNetLMHeadModel): The instance of the XLNetLMHeadModel class.
        new_embeddings (tensor): The new output embeddings to be set for the model.
            It should be a tensor of the appropriate shape and type.

    Returns:
        None.

    Raises:
        TypeError: If the new_embeddings parameter is not of type tensor.
        ValueError: If the new_embeddings parameter does not meet the required shape or type constraints.
    """
    self.lm_loss = new_embeddings

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLMHeadModelOutput dataclass

Bases: ModelOutput

Output type of [XLNetLMHeadModel].

PARAMETER DESCRIPTION
logits

Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

num_predict corresponds to target_mapping.shape[1]. If target_mapping is None, then num_predict corresponds to sequence_length.

TYPE: `torch.FloatTensor` of shape `(batch_size, num_predict, config.vocab_size)` DEFAULT: None

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
@dataclass
class XLNetLMHeadModelOutput(ModelOutput):
    """
    Output type of [`XLNetLMHeadModel`].

    Args:
        loss (`torch.FloatTensor` of shape *(1,)*, *optional*, returned when `labels` is provided)
            Language modeling loss (for next-token prediction).
        logits (`torch.FloatTensor` of shape `(batch_size, num_predict, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

            `num_predict` corresponds to `target_mapping.shape[1]`. If `target_mapping` is `None`, then `num_predict`
            corresponds to `sequence_length`.
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    loss: Optional[mindspore.Tensor] = None
    logits: mindspore.Tensor = None
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLayer

Bases: Cell

Represents a layer of the XLNet model. This class includes methods for initializing the layer, constructing the layer's output, and applying chunking to the forward pass.

This class inherits from the nn.Cell class.

ATTRIBUTE DESCRIPTION
rel_attn

XLNetRelativeAttention The XLNetRelativeAttention instance for relative attention computation.
ff

XLNetFeedForward The XLNetFeedForward instance for feed-forward computation.
dropout

nn.Dropout The dropout instance for regularization.
chunk_size_feed_forward

int The chunk size for feed-forward computation.
seq_len_dim

int The sequence length dimension.
METHOD DESCRIPTION
__init__

Initializes the XLNetLayer with the provided configuration.
construct

Constructs the output of the XLNetLayer based on the provided inputs and optional arguments.
ff_chunk

Applies chunking to the forward pass for the provided output_x.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
class XLNetLayer(nn.Cell):

    """
    Represents a layer of the XLNet model. This class includes methods for initializing the layer,
    constructing the layer's output, and applying chunking to the forward pass.

    This class inherits from the nn.Cell class.

    Attributes:
        rel_attn: XLNetRelativeAttention
            The XLNetRelativeAttention instance for relative attention computation.
        ff: XLNetFeedForward
            The XLNetFeedForward instance for feed-forward computation.
        dropout: nn.Dropout
            The dropout instance for regularization.
        chunk_size_feed_forward: int
            The chunk size for feed-forward computation.
        seq_len_dim: int
            The sequence length dimension.

    Methods:
        __init__:
            Initializes the XLNetLayer with the provided configuration.

        construct:
            Constructs the output of the XLNetLayer based on the provided inputs and optional arguments.

        ff_chunk:
            Applies chunking to the forward pass for the provided output_x.
    """
    def __init__(self, config):
        """
        Initializes an instance of the XLNetLayer class.

        Args:
            self: The instance of the XLNetLayer class.
            config: A configuration object containing parameters for the XLNetLayer initialization.

        Returns:
            None.

        Raises:
            None.
        """
        super().__init__()
        self.rel_attn = XLNetRelativeAttention(config)
        self.ff = XLNetFeedForward(config)
        self.dropout = nn.Dropout(p=config.dropout)
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1

    def construct(
            self,
            output_h,
            output_g,
            attn_mask_h,
            attn_mask_g,
            r,
            seg_mat,
            mems=None,
            target_mapping=None,
            head_mask=None,
            output_attentions=False,
    ):
        """
        This method constructs the XLNet layer.

        Args:
            self: The instance of the XLNetLayer class.
            output_h (tensor): The output tensor from the previous layer for the current head.
            output_g (tensor): The output tensor from the previous layer for the global context.
            attn_mask_h (tensor): The attention mask for the current head.
            attn_mask_g (tensor): The attention mask for the global context.
            r (int): The number of attention heads.
            seg_mat (tensor): The segment matrix specifying the segment for each token.
            mems (tensor, optional): The memory tensor. Defaults to None.
            target_mapping (tensor, optional): The target mapping tensor. Defaults to None.
            head_mask (tensor, optional): The head mask tensor. Defaults to None.
            output_attentions (bool, optional): Controls whether to output attentions. Defaults to False.

        Returns:
            tuple: A tuple containing the output tensors for the current head and the global context,
                and any additional outputs.

        Raises:
            ValueError: If the dimensions of input tensors are not compatible.
            RuntimeError: If there is a runtime issue during the execution of the method.
            TypeError: If the input types are not as expected.
        """
        outputs = self.rel_attn(
            output_h,
            output_g,
            attn_mask_h,
            attn_mask_g,
            r,
            seg_mat,
            mems=mems,
            target_mapping=target_mapping,
            head_mask=head_mask,
            output_attentions=output_attentions,
        )
        output_h, output_g = outputs[:2]

        if output_g is not None:
            output_g = apply_chunking_to_forward(
                self.ff_chunk, self.chunk_size_feed_forward, self.seq_len_dim, output_g
            )
        output_h = apply_chunking_to_forward(self.ff_chunk, self.chunk_size_feed_forward, self.seq_len_dim, output_h)

        outputs = (output_h, output_g) + outputs[2:]
        return outputs

    def ff_chunk(self, output_x):
        """
        Performs a forward pass through the XLNetLayer for a given input chunk.

        Args:
            self (XLNetLayer): An instance of the XLNetLayer class.
            output_x: The input chunk to be processed. It should be a tensor.

        Returns:
            None.

        Raises:
            None.
        """
        output_x = self.ff(output_x)
        return output_x

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLayer.__init__(config)

Initializes an instance of the XLNetLayer class.

PARAMETER DESCRIPTION
self

The instance of the XLNetLayer class.
config

A configuration object containing parameters for the XLNetLayer initialization.
RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
def __init__(self, config):
    """
    Initializes an instance of the XLNetLayer class.

    Args:
        self: The instance of the XLNetLayer class.
        config: A configuration object containing parameters for the XLNetLayer initialization.

    Returns:
        None.

    Raises:
        None.
    """
    super().__init__()
    self.rel_attn = XLNetRelativeAttention(config)
    self.ff = XLNetFeedForward(config)
    self.dropout = nn.Dropout(p=config.dropout)
    self.chunk_size_feed_forward = config.chunk_size_feed_forward
    self.seq_len_dim = 1

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLayer.construct(output_h, output_g, attn_mask_h, attn_mask_g, r, seg_mat, mems=None, target_mapping=None, head_mask=None, output_attentions=False)

This method constructs the XLNet layer.

PARAMETER DESCRIPTION
self

The instance of the XLNetLayer class.
output_h

The output tensor from the previous layer for the current head.

TYPE: tensor

output_g

The output tensor from the previous layer for the global context.

TYPE: tensor

attn_mask_h

The attention mask for the current head.

TYPE: tensor

attn_mask_g

The attention mask for the global context.

TYPE: tensor

r

The number of attention heads.

TYPE: int

seg_mat

The segment matrix specifying the segment for each token.

TYPE: tensor

mems

The memory tensor. Defaults to None.

TYPE: tensor DEFAULT: None

target_mapping

The target mapping tensor. Defaults to None.

TYPE: tensor DEFAULT: None

head_mask

The head mask tensor. Defaults to None.

TYPE: tensor DEFAULT: None

output_attentions

Controls whether to output attentions. Defaults to False.

TYPE: bool DEFAULT: False

RETURNS DESCRIPTION
tuple

A tuple containing the output tensors for the current head and the global context, and any additional outputs.
RAISES DESCRIPTION
ValueError

If the dimensions of input tensors are not compatible.
RuntimeError

If there is a runtime issue during the execution of the method.
TypeError

If the input types are not as expected.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
def construct(
        self,
        output_h,
        output_g,
        attn_mask_h,
        attn_mask_g,
        r,
        seg_mat,
        mems=None,
        target_mapping=None,
        head_mask=None,
        output_attentions=False,
):
    """
    This method constructs the XLNet layer.

    Args:
        self: The instance of the XLNetLayer class.
        output_h (tensor): The output tensor from the previous layer for the current head.
        output_g (tensor): The output tensor from the previous layer for the global context.
        attn_mask_h (tensor): The attention mask for the current head.
        attn_mask_g (tensor): The attention mask for the global context.
        r (int): The number of attention heads.
        seg_mat (tensor): The segment matrix specifying the segment for each token.
        mems (tensor, optional): The memory tensor. Defaults to None.
        target_mapping (tensor, optional): The target mapping tensor. Defaults to None.
        head_mask (tensor, optional): The head mask tensor. Defaults to None.
        output_attentions (bool, optional): Controls whether to output attentions. Defaults to False.

    Returns:
        tuple: A tuple containing the output tensors for the current head and the global context,
            and any additional outputs.

    Raises:
        ValueError: If the dimensions of input tensors are not compatible.
        RuntimeError: If there is a runtime issue during the execution of the method.
        TypeError: If the input types are not as expected.
    """
    outputs = self.rel_attn(
        output_h,
        output_g,
        attn_mask_h,
        attn_mask_g,
        r,
        seg_mat,
        mems=mems,
        target_mapping=target_mapping,
        head_mask=head_mask,
        output_attentions=output_attentions,
    )
    output_h, output_g = outputs[:2]

    if output_g is not None:
        output_g = apply_chunking_to_forward(
            self.ff_chunk, self.chunk_size_feed_forward, self.seq_len_dim, output_g
        )
    output_h = apply_chunking_to_forward(self.ff_chunk, self.chunk_size_feed_forward, self.seq_len_dim, output_h)

    outputs = (output_h, output_g) + outputs[2:]
    return outputs

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetLayer.ff_chunk(output_x)

Performs a forward pass through the XLNetLayer for a given input chunk.

PARAMETER DESCRIPTION
self

An instance of the XLNetLayer class.

TYPE: XLNetLayer

output_x

The input chunk to be processed. It should be a tensor.
RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
def ff_chunk(self, output_x):
    """
    Performs a forward pass through the XLNetLayer for a given input chunk.

    Args:
        self (XLNetLayer): An instance of the XLNetLayer class.
        output_x: The input chunk to be processed. It should be a tensor.

    Returns:
        None.

    Raises:
        None.
    """
    output_x = self.ff(output_x)
    return output_x

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel

Bases: XLNetPreTrainedModel

The XLNetModel class represents a model for XLNet, which is a type of pre-trained model for natural language processing. It inherits from the XLNetPreTrainedModel class and provides methods for initializing the model, creating attention masks, caching memory, and constructing the model for inference. The class also includes methods for managing input embeddings, positional embeddings, and relative positional encoding.

The class includes methods for creating attention masks, caching memory, and constructing the model for inference. It also provides functionality for managing input embeddings, positional embeddings, and relative positional encoding. The class methods are designed to handle various input parameters and configurations for fine-tuning and using the XLNet model for specific NLP tasks. The class is designed to be flexible and efficient for handling different use cases and scenarios.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
class XLNetModel(XLNetPreTrainedModel):

    """
    The XLNetModel class represents a model for XLNet, which is a type of pre-trained model for natural language 
    processing. It inherits from the XLNetPreTrainedModel class and provides methods for initializing the model, 
    creating attention masks, caching memory, and constructing the model for inference. The class also includes 
    methods for managing input embeddings, positional embeddings, and relative positional encoding.

    The class includes methods for creating attention masks, caching memory, and constructing the model for inference. 
    It also provides functionality for managing input embeddings, positional embeddings, and relative positional 
    encoding. The class methods are designed to handle various input parameters and configurations for fine-tuning and 
    using the XLNet model for specific NLP tasks. The class is designed to be flexible and efficient for handling 
    different use cases and scenarios.
    """
    def __init__(self, config):
        """
        This method initializes an instance of the XLNetModel class with the provided configuration.

        Args:
            self: The instance of the XLNetModel class.
            config: 
                A configuration object containing the following parameters:

                - mem_len (int): The length of the memory.
                - reuse_len (int): The length of the segment that can be reused.
                - d_model (int): The dimension of the model.
                - same_length (bool): A flag indicating whether the segments have the same length.
                - attn_type (str): The type of attention mechanism to be used.
                - bi_data (bool): A flag indicating whether the input data is bidirectional.
                - clamp_len (int): The maximum length of the segments.
                - n_layer (int): The number of layers in the model.
                - vocab_size (int): The size of the vocabulary for word embeddings.
                - dropout (float): The dropout rate.

        Returns:
            None.

        Raises:
            ValueError: If the provided configuration is invalid or incomplete.
            TypeError: If the data types of the configuration parameters are not as expected.
            RuntimeError: If an error occurs during the initialization process.
        """
        super().__init__(config)

        self.mem_len = config.mem_len
        self.reuse_len = config.reuse_len
        self.d_model = config.d_model
        self.same_length = config.same_length
        self.attn_type = config.attn_type
        self.bi_data = config.bi_data
        self.clamp_len = config.clamp_len
        self.n_layer = config.n_layer

        self.word_embedding = nn.Embedding(config.vocab_size, config.d_model)
        self.mask_emb = mindspore.Parameter(ops.zeros((1, 1, config.d_model),dtype=mindspore.float32))
        self.layer = nn.CellList([XLNetLayer(config) for _ in range(config.n_layer)])
        self.dropout = nn.Dropout(p=config.dropout)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        """
        This method retrieves the input embeddings from the XLNetModel.

        Args:
            self (XLNetModel): The instance of the XLNetModel class.
                The self parameter is required to access the word_embedding attribute.

        Returns:
            None.

        Raises:
            None.
        """
        return self.word_embedding

    def set_input_embeddings(self, new_embeddings):
        """
        Method to set new input embeddings for the XLNetModel.

        Args:
            self (XLNetModel): The instance of the XLNetModel class.
                This parameter is a reference to the current XLNetModel instance where the embeddings will be set.
            new_embeddings (any): The new input embeddings to be assigned to the XLNetModel.
                This parameter represents the new embeddings that will replace the existing word embeddings in the model.

        Returns:
            None.

        Raises:
            None.
        """
        self.word_embedding = new_embeddings

    def _prune_heads(self, heads_to_prune):
        """
        This method `_prune_heads` is a part of the `XLNetModel` class and is used to prune specific heads from the model.

        Args:
            self (XLNetModel): The instance of the XLNetModel class.
            heads_to_prune (int[]): A list of integers representing the indices of the heads to be pruned from the model.
                The indices should be within the valid range of heads for the model.

        Returns:
            None: This method does not return any value.
                It performs the operation in place by pruning the specified heads from the model.

        Raises:
            NotImplementedError: If this method is called directly,
                it raises a NotImplementedError as it should be implemented in a subclass.
        """
        raise NotImplementedError

    def create_mask(self, qlen, mlen):
        """
        Creates causal attention mask. Float mask where 1.0 indicates masked, 0.0 indicates not-masked.

        Args:
            qlen: Sequence length
            mlen: Mask length

        ::

                  same_length=False: same_length=True: <mlen > < qlen > <mlen > < qlen >
               ^ [0 0 0 0 0 1 1 1 1] [0 0 0 0 0 1 1 1 1]
                 [0 0 0 0 0 0 1 1 1] [1 0 0 0 0 0 1 1 1]
            qlen [0 0 0 0 0 0 0 1 1] [1 1 0 0 0 0 0 1 1]
                 [0 0 0 0 0 0 0 0 1] [1 1 1 0 0 0 0 0 1]
               v [0 0 0 0 0 0 0 0 0] [1 1 1 1 0 0 0 0 0]

        """
        mask = ops.ones((qlen, qlen + mlen))
        if self.same_length:
            mask_lo = mask[:, :qlen].tril(-1)
            mask.triu_(mlen + 1)
            mask[:, :qlen] += mask_lo
        else:
            mask.triu_(mlen + 1)

        return mask

    def cache_mem(self, curr_out, prev_mem):
        """
        Caches memory for the XLNetModel.

        Args:
            self (XLNetModel): The instance of the XLNetModel class.
            curr_out (Tensor): The current output tensor.
            prev_mem (Tensor): The previous memory tensor.

        Returns:
            None.

        Raises:
            None.

        """
        # cache hidden states into memory.
        if self.reuse_len is not None and self.reuse_len > 0:
            curr_out = curr_out[: self.reuse_len]

        if self.mem_len is None or self.mem_len == 0:
            # If `use_mems` is active but no `mem_len` is defined, the model behaves like GPT-2 at inference time
            # and returns all of the past and current hidden states.
            cutoff = 0
        else:
            # If `use_mems` is active and `mem_len` is defined, the model returns the last `mem_len` hidden
            # states. This is the preferred setting for training and long-form generation.
            cutoff = -self.mem_len
        if prev_mem is None:
            # if `use_mems` is active and `mem_len` is defined, the model
            new_mem = curr_out[cutoff:]
        else:
            new_mem = ops.cat([prev_mem, curr_out], axis=0)[cutoff:]

        return new_mem

    @staticmethod
    def positional_embedding(pos_seq, inv_freq, bsz=None):
        """
        This method is a static method in the class 'XLNetModel' and is used to generate positional embeddings for input sequences.

        Args:
            pos_seq (torch.Tensor):
                A tensor containing the positional sequence.

                - Type: torch.Tensor
                - Purpose: This tensor represents the positions of the input tokens in the sequence.
                - Restrictions: None

            inv_freq (torch.Tensor):
                A tensor containing the inverse frequency values.

                - Type: torch.Tensor
                - Purpose: This tensor represents the inverse frequency values to be used in the positional
                embedding calculation.
                - Restrictions: None

            bsz (int, optional):
                An optional parameter representing the batch size.

                - Type: int
                - Purpose: This parameter is used to expand the positional embeddings tensor if provided.
                - Restrictions: None

        Returns:
            torch.Tensor:
                A tensor containing the positional embeddings.

                - Type: torch.Tensor
                - Purpose: This tensor represents the positional embeddings for the input sequence.

        Raises:
            None
        """
        sinusoid_inp = ops.einsum("i,d->id", pos_seq, inv_freq)
        pos_emb = ops.cat([ops.sin(sinusoid_inp), ops.cos(sinusoid_inp)], axis=-1)
        pos_emb = pos_emb[:, None, :]

        if bsz is not None:
            pos_emb = pos_emb.expand(-1, bsz, -1)

        return pos_emb

    def relative_positional_encoding(self, qlen, klen, bsz=None):
        """
        Encodes relative positional information for the XLNetModel.

        Args:
            self (XLNetModel): The instance of the XLNetModel class.
            qlen (int): The length of the query sequence.
            klen (int): The length of the key sequence.
            bsz (int, optional): The batch size. Defaults to None.

        Returns:
            None

        Raises:
            ValueError: If the `attn_type` is not 'bi' or 'uni'.

        """
        # create relative positional encoding.
        freq_seq = ops.arange(0, self.d_model, 2.0, dtype=mindspore.int64).float()
        inv_freq = 1 / ops.pow(10000, (freq_seq / self.d_model))

        if self.attn_type == "bi":
            # beg, end = klen - 1, -qlen
            beg, end = klen, -qlen
        elif self.attn_type == "uni":
            # beg, end = klen - 1, -1
            beg, end = klen, -1
        else:
            raise ValueError(f"Unknown `attn_type` {self.attn_type}.")

        if self.bi_data:
            fwd_pos_seq = ops.arange(beg, end, -1.0, dtype=mindspore.int64).float()
            bwd_pos_seq = ops.arange(-beg, -end, 1.0, dtype=mindspore.int64).float()

            if self.clamp_len > 0:
                fwd_pos_seq = fwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)
                bwd_pos_seq = bwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)

            if bsz is not None:
                fwd_pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq, bsz // 2)
                bwd_pos_emb = self.positional_embedding(bwd_pos_seq, inv_freq, bsz // 2)
            else:
                fwd_pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq)
                bwd_pos_emb = self.positional_embedding(bwd_pos_seq, inv_freq)

            pos_emb = ops.cat([fwd_pos_emb, bwd_pos_emb], axis=1)
        else:
            fwd_pos_seq = ops.arange(beg, end, -1.0, dtype=mindspore.int64).float()
            if self.clamp_len > 0:
                fwd_pos_seq = fwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)
            pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq, bsz)

        return pos_emb

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            mems: Optional[mindspore.Tensor] = None,
            perm_mask: Optional[mindspore.Tensor] = None,
            target_mapping: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            input_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            use_mems: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            **kwargs,  # delete after depreciation warning is removed
    ) -> Union[Tuple, XLNetModelOutput]:
        """
        Args:
            self: The instance of the XLNetModel class.
            input_ids (Optional[mindspore.Tensor]): The input tensor containing the token IDs. Default is None.
            attention_mask (Optional[mindspore.Tensor]): The attention mask tensor to avoid attending to padding tokens.
                Default is None.
            mems (Optional[mindspore.Tensor]): The memory tensor for caching previous hidden states. Default is None.
            perm_mask (Optional[mindspore.Tensor]): The permutation mask tensor for partial attention over sequence.
                Default is None.
            target_mapping (Optional[mindspore.Tensor]): The target mapping tensor for masked language modeling.
                Default is None.
            token_type_ids (Optional[mindspore.Tensor]): The tensor containing token type IDs for differentiating
                sequences. Default is None.
            input_mask (Optional[mindspore.Tensor]): The input mask tensor indicating padding tokens. Default is None.
            head_mask (Optional[mindspore.Tensor]): The mask tensor for controlling the attention heads. Default is None.
            inputs_embeds (Optional[mindspore.Tensor]): The tensor containing precomputed embeddings. Default is None.
            use_mems (Optional[bool]): Flag indicating whether to use memory for caching. Default is None.
            output_attentions (Optional[bool]): Flag indicating whether to output attention weights. Default is None.
            output_hidden_states (Optional[bool]): Flag indicating whether to output hidden states of all layers.
                Default is None.
            return_dict (Optional[bool]): Flag indicating whether to return output as a dict. Default is None.

        Returns:
            Union[Tuple, XLNetModelOutput]: The output of the XLNetModel construct method, which includes the last
                hidden state, memory tensors, hidden states of all layers, and attention weights.

        Raises:
            ValueError: Raised if both input_ids and inputs_embeds are specified simultaneously, or if neither
                input_ids nor inputs_embeds are specified.
            FutureWarning: Raised when the 'use_cache' argument is deprecated. Use 'use_mems' instead.
            ValueError: Raised if an unsupported attention type is encountered.
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if "use_cache" in kwargs:
            warnings.warn(
                "The `use_cache` argument is deprecated and will be removed in a future version, use `use_mems`"
                " instead.",
                FutureWarning,
            )
            use_mems = kwargs["use_cache"]

        if self.training:
            use_mems = use_mems if use_mems is not None else self.config.use_mems_train
        else:
            use_mems = use_mems if use_mems is not None else self.config.use_mems_eval

        # the original code for XLNet uses shapes [len, bsz] with the batch dimension at the end
        # but we want a unified interface in the library with the batch size on the first dimension
        # so we move here the first dimension (batch) to the end
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            input_ids = input_ids.swapaxes(1, 0)
            qlen, bsz = input_ids.shape[0], input_ids.shape[1]
        elif inputs_embeds is not None:
            inputs_embeds = inputs_embeds.transpose(1, 0, 2)
            qlen, bsz = inputs_embeds.shape[0], inputs_embeds.shape[1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        token_type_ids = token_type_ids.swapaxes(1, 0) if token_type_ids is not None else None
        input_mask = input_mask.swapaxes(1, 0) if input_mask is not None else None
        attention_mask = attention_mask.swapaxes(1, 0) if attention_mask is not None else None
        perm_mask = perm_mask.permute(1, 2, 0) if perm_mask is not None else None
        target_mapping = target_mapping.permute(1, 2, 0) if target_mapping is not None else None

        mlen = mems[0].shape[0] if mems is not None and mems[0] is not None else 0
        klen = mlen + qlen

        dtype_float = self.dtype

        # Attention mask
        # causal attention mask
        if self.attn_type == "uni":
            attn_mask = self.create_mask(qlen, mlen)
            attn_mask = attn_mask[:, :, None, None]
        elif self.attn_type == "bi":
            attn_mask = None
        else:
            raise ValueError(f"Unsupported attention type: {self.attn_type}")

        # data mask: input mask & perm mask
        assert input_mask is None or attention_mask is None, "You can only use one of input_mask (uses 1 for padding) "
        "or attention_mask (uses 0 for padding, added for compatibility with BERT). Please choose one."
        if input_mask is None and attention_mask is not None:
            input_mask = 1.0 - attention_mask
        if input_mask is not None and perm_mask is not None:
            data_mask = input_mask[None] + perm_mask
        elif input_mask is not None and perm_mask is None:
            data_mask = input_mask[None]
        elif input_mask is None and perm_mask is not None:
            data_mask = perm_mask
        else:
            data_mask = None

        if data_mask is not None:
            # all mems can be attended to
            if mlen > 0:
                mems_mask = ops.zeros((data_mask.shape[0], mlen, bsz)).to(dtype=data_mask.dtype)
                data_mask = ops.cat([mems_mask, data_mask], axis=1)
            if attn_mask is None:
                attn_mask = data_mask[:, :, :, None]
            else:
                attn_mask += data_mask[:, :, :, None]

        if attn_mask is not None:
            attn_mask = (attn_mask > 0).to(dtype_float)

        if attn_mask is not None:
            non_tgt_mask = -ops.eye(qlen).to(dtype=attn_mask.dtype)
            if mlen > 0:
                non_tgt_mask = ops.cat([ops.zeros((qlen, mlen)).to(dtype=attn_mask.dtype), non_tgt_mask],
                                                 axis=-1)
            non_tgt_mask = ((attn_mask + non_tgt_mask[:, :, None, None]) > 0).to(dtype=attn_mask.dtype)
        else:
            non_tgt_mask = None

        # Word embeddings and prepare h & g hidden states
        if inputs_embeds is not None:
            word_emb_k = inputs_embeds
        else:
            word_emb_k = self.word_embedding(input_ids)
        output_h = self.dropout(word_emb_k)
        if target_mapping is not None:
            word_emb_q = self.mask_emb.expand(target_mapping.shape[0], bsz, -1)
            # else:  # We removed the inp_q input which was same as target mapping
            #     inp_q_ext = inp_q[:, :, None]
            #     word_emb_q = inp_q_ext * self.mask_emb + (1 - inp_q_ext) * word_emb_k
            output_g = self.dropout(word_emb_q)
        else:
            output_g = None

        # Segment embedding
        if token_type_ids is not None:
            # Convert `token_type_ids` to one-hot `seg_mat`
            if mlen > 0:
                mem_pad = ops.zeros((mlen, bsz), dtype=mindspore.int64)
                cat_ids = ops.cat([mem_pad, token_type_ids], axis=0)
            else:
                cat_ids = token_type_ids

            # `1` indicates not in the same segment [qlen x klen x bsz]
            seg_mat = (token_type_ids[:, None] != cat_ids[None, :]).long()
            seg_mat = ops.one_hot(seg_mat, depth=2).to(dtype_float)
        else:
            seg_mat = None

        # Positional encoding
        pos_emb = self.relative_positional_encoding(qlen, klen, bsz=bsz)
        pos_emb = self.dropout(pos_emb)

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads] (a head_mask for each layer)
        # and head_mask is converted to shape [num_hidden_layers x qlen x klen x bsz x n_head]
        if head_mask is not None:
            if head_mask.dim() == 1:
                head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(0).unsqueeze(0)
                head_mask = head_mask.expand(self.n_layer, -1, -1, -1, -1)
            elif head_mask.dim() == 2:
                head_mask = head_mask.unsqueeze(1).unsqueeze(1).unsqueeze(1)
            head_mask_dtype = next(iter(self.parameters_dict().items()))[1].dtype
            head_mask = head_mask.to(dtype=head_mask_dtype)
        else:
            head_mask = [None] * self.n_layer

        new_mems = ()
        if mems is None:
            mems = [None] * len(self.layer)

        attentions = [] if output_attentions else None
        hidden_states = [] if output_hidden_states else None
        for i, layer_module in enumerate(self.layer):
            if use_mems:
                # cache new mems
                new_mems = new_mems + (self.cache_mem(output_h, mems[i]),)
            if output_hidden_states:
                hidden_states.append((output_h, output_g) if output_g is not None else output_h)

            outputs = layer_module(
                output_h,
                output_g,
                attn_mask_h=non_tgt_mask,
                attn_mask_g=attn_mask,
                r=pos_emb,
                seg_mat=seg_mat,
                mems=mems[i],
                target_mapping=target_mapping,
                head_mask=head_mask[i],
                output_attentions=output_attentions,
            )
            output_h, output_g = outputs[:2]
            if output_attentions:
                attentions.append(outputs[2])

        # Add last hidden state
        if output_hidden_states:
            hidden_states.append((output_h, output_g) if output_g is not None else output_h)

        output = self.dropout(output_g if output_g is not None else output_h)

        # Prepare outputs, we swapaxes back here to shape [bsz, len, hidden_dim] (cf. beginning of forward() method)
        output = output.permute(1, 0, 2)

        if not use_mems:
            new_mems = None

        if output_hidden_states:
            if output_g is not None:
                hidden_states = tuple(h.permute(1, 0, 2) for hs in hidden_states for h in hs)
            else:
                hidden_states = tuple(hs.permute(1, 0, 2) for hs in hidden_states)

        if output_attentions:
            if target_mapping is not None:
                # when target_mapping is provided, there are 2-tuple of attentions
                attentions = tuple(
                    tuple(att_stream.permute(2, 3, 0, 1) for att_stream in t) for t in attentions
                )
            else:
                attentions = tuple(t.permute(2, 3, 0, 1) for t in attentions)

        if not return_dict:
            return tuple(v for v in [output, new_mems, hidden_states, attentions] if v is not None)

        return XLNetModelOutput(
            last_hidden_state=output, mems=new_mems, hidden_states=hidden_states, attentions=attentions
        )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.__init__(config)

This method initializes an instance of the XLNetModel class with the provided configuration.

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class.
config

A configuration object containing the following parameters:

  • mem_len (int): The length of the memory.
  • reuse_len (int): The length of the segment that can be reused.
  • d_model (int): The dimension of the model.
  • same_length (bool): A flag indicating whether the segments have the same length.
  • attn_type (str): The type of attention mechanism to be used.
  • bi_data (bool): A flag indicating whether the input data is bidirectional.
  • clamp_len (int): The maximum length of the segments.
  • n_layer (int): The number of layers in the model.
  • vocab_size (int): The size of the vocabulary for word embeddings.
  • dropout (float): The dropout rate.
RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
ValueError

If the provided configuration is invalid or incomplete.
TypeError

If the data types of the configuration parameters are not as expected.
RuntimeError

If an error occurs during the initialization process.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
def __init__(self, config):
    """
    This method initializes an instance of the XLNetModel class with the provided configuration.

    Args:
        self: The instance of the XLNetModel class.
        config: 
            A configuration object containing the following parameters:

            - mem_len (int): The length of the memory.
            - reuse_len (int): The length of the segment that can be reused.
            - d_model (int): The dimension of the model.
            - same_length (bool): A flag indicating whether the segments have the same length.
            - attn_type (str): The type of attention mechanism to be used.
            - bi_data (bool): A flag indicating whether the input data is bidirectional.
            - clamp_len (int): The maximum length of the segments.
            - n_layer (int): The number of layers in the model.
            - vocab_size (int): The size of the vocabulary for word embeddings.
            - dropout (float): The dropout rate.

    Returns:
        None.

    Raises:
        ValueError: If the provided configuration is invalid or incomplete.
        TypeError: If the data types of the configuration parameters are not as expected.
        RuntimeError: If an error occurs during the initialization process.
    """
    super().__init__(config)

    self.mem_len = config.mem_len
    self.reuse_len = config.reuse_len
    self.d_model = config.d_model
    self.same_length = config.same_length
    self.attn_type = config.attn_type
    self.bi_data = config.bi_data
    self.clamp_len = config.clamp_len
    self.n_layer = config.n_layer

    self.word_embedding = nn.Embedding(config.vocab_size, config.d_model)
    self.mask_emb = mindspore.Parameter(ops.zeros((1, 1, config.d_model),dtype=mindspore.float32))
    self.layer = nn.CellList([XLNetLayer(config) for _ in range(config.n_layer)])
    self.dropout = nn.Dropout(p=config.dropout)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.cache_mem(curr_out, prev_mem)

Caches memory for the XLNetModel.

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class.

TYPE: XLNetModel

curr_out

The current output tensor.

TYPE: Tensor

prev_mem

The previous memory tensor.

TYPE: Tensor

RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
def cache_mem(self, curr_out, prev_mem):
    """
    Caches memory for the XLNetModel.

    Args:
        self (XLNetModel): The instance of the XLNetModel class.
        curr_out (Tensor): The current output tensor.
        prev_mem (Tensor): The previous memory tensor.

    Returns:
        None.

    Raises:
        None.

    """
    # cache hidden states into memory.
    if self.reuse_len is not None and self.reuse_len > 0:
        curr_out = curr_out[: self.reuse_len]

    if self.mem_len is None or self.mem_len == 0:
        # If `use_mems` is active but no `mem_len` is defined, the model behaves like GPT-2 at inference time
        # and returns all of the past and current hidden states.
        cutoff = 0
    else:
        # If `use_mems` is active and `mem_len` is defined, the model returns the last `mem_len` hidden
        # states. This is the preferred setting for training and long-form generation.
        cutoff = -self.mem_len
    if prev_mem is None:
        # if `use_mems` is active and `mem_len` is defined, the model
        new_mem = curr_out[cutoff:]
    else:
        new_mem = ops.cat([prev_mem, curr_out], axis=0)[cutoff:]

    return new_mem

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.construct(input_ids=None, attention_mask=None, mems=None, perm_mask=None, target_mapping=None, token_type_ids=None, input_mask=None, head_mask=None, inputs_embeds=None, use_mems=None, output_attentions=None, output_hidden_states=None, return_dict=None, **kwargs)

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class.
input_ids

The input tensor containing the token IDs. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

attention_mask

The attention mask tensor to avoid attending to padding tokens. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

mems

The memory tensor for caching previous hidden states. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

perm_mask

The permutation mask tensor for partial attention over sequence. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

target_mapping

The target mapping tensor for masked language modeling. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

token_type_ids

The tensor containing token type IDs for differentiating sequences. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

input_mask

The input mask tensor indicating padding tokens. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

head_mask

The mask tensor for controlling the attention heads. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

inputs_embeds

The tensor containing precomputed embeddings. Default is None.

TYPE: Optional[Tensor] DEFAULT: None

use_mems

Flag indicating whether to use memory for caching. Default is None.

TYPE: Optional[bool] DEFAULT: None

output_attentions

Flag indicating whether to output attention weights. Default is None.

TYPE: Optional[bool] DEFAULT: None

output_hidden_states

Flag indicating whether to output hidden states of all layers. Default is None.

TYPE: Optional[bool] DEFAULT: None

return_dict

Flag indicating whether to return output as a dict. Default is None.

TYPE: Optional[bool] DEFAULT: None

RETURNS DESCRIPTION
Union[Tuple, XLNetModelOutput]

Union[Tuple, XLNetModelOutput]: The output of the XLNetModel construct method, which includes the last hidden state, memory tensors, hidden states of all layers, and attention weights.
RAISES DESCRIPTION
ValueError

Raised if both input_ids and inputs_embeds are specified simultaneously, or if neither input_ids nor inputs_embeds are specified.
FutureWarning

Raised when the 'use_cache' argument is deprecated. Use 'use_mems' instead.
ValueError

Raised if an unsupported attention type is encountered.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
def construct(
        self,
        input_ids: Optional[mindspore.Tensor] = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        mems: Optional[mindspore.Tensor] = None,
        perm_mask: Optional[mindspore.Tensor] = None,
        target_mapping: Optional[mindspore.Tensor] = None,
        token_type_ids: Optional[mindspore.Tensor] = None,
        input_mask: Optional[mindspore.Tensor] = None,
        head_mask: Optional[mindspore.Tensor] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        use_mems: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,  # delete after depreciation warning is removed
) -> Union[Tuple, XLNetModelOutput]:
    """
    Args:
        self: The instance of the XLNetModel class.
        input_ids (Optional[mindspore.Tensor]): The input tensor containing the token IDs. Default is None.
        attention_mask (Optional[mindspore.Tensor]): The attention mask tensor to avoid attending to padding tokens.
            Default is None.
        mems (Optional[mindspore.Tensor]): The memory tensor for caching previous hidden states. Default is None.
        perm_mask (Optional[mindspore.Tensor]): The permutation mask tensor for partial attention over sequence.
            Default is None.
        target_mapping (Optional[mindspore.Tensor]): The target mapping tensor for masked language modeling.
            Default is None.
        token_type_ids (Optional[mindspore.Tensor]): The tensor containing token type IDs for differentiating
            sequences. Default is None.
        input_mask (Optional[mindspore.Tensor]): The input mask tensor indicating padding tokens. Default is None.
        head_mask (Optional[mindspore.Tensor]): The mask tensor for controlling the attention heads. Default is None.
        inputs_embeds (Optional[mindspore.Tensor]): The tensor containing precomputed embeddings. Default is None.
        use_mems (Optional[bool]): Flag indicating whether to use memory for caching. Default is None.
        output_attentions (Optional[bool]): Flag indicating whether to output attention weights. Default is None.
        output_hidden_states (Optional[bool]): Flag indicating whether to output hidden states of all layers.
            Default is None.
        return_dict (Optional[bool]): Flag indicating whether to return output as a dict. Default is None.

    Returns:
        Union[Tuple, XLNetModelOutput]: The output of the XLNetModel construct method, which includes the last
            hidden state, memory tensors, hidden states of all layers, and attention weights.

    Raises:
        ValueError: Raised if both input_ids and inputs_embeds are specified simultaneously, or if neither
            input_ids nor inputs_embeds are specified.
        FutureWarning: Raised when the 'use_cache' argument is deprecated. Use 'use_mems' instead.
        ValueError: Raised if an unsupported attention type is encountered.
    """
    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
    output_hidden_states = (
        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
    )
    return_dict = return_dict if return_dict is not None else self.config.use_return_dict

    if "use_cache" in kwargs:
        warnings.warn(
            "The `use_cache` argument is deprecated and will be removed in a future version, use `use_mems`"
            " instead.",
            FutureWarning,
        )
        use_mems = kwargs["use_cache"]

    if self.training:
        use_mems = use_mems if use_mems is not None else self.config.use_mems_train
    else:
        use_mems = use_mems if use_mems is not None else self.config.use_mems_eval

    # the original code for XLNet uses shapes [len, bsz] with the batch dimension at the end
    # but we want a unified interface in the library with the batch size on the first dimension
    # so we move here the first dimension (batch) to the end
    if input_ids is not None and inputs_embeds is not None:
        raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
    elif input_ids is not None:
        input_ids = input_ids.swapaxes(1, 0)
        qlen, bsz = input_ids.shape[0], input_ids.shape[1]
    elif inputs_embeds is not None:
        inputs_embeds = inputs_embeds.transpose(1, 0, 2)
        qlen, bsz = inputs_embeds.shape[0], inputs_embeds.shape[1]
    else:
        raise ValueError("You have to specify either input_ids or inputs_embeds")

    token_type_ids = token_type_ids.swapaxes(1, 0) if token_type_ids is not None else None
    input_mask = input_mask.swapaxes(1, 0) if input_mask is not None else None
    attention_mask = attention_mask.swapaxes(1, 0) if attention_mask is not None else None
    perm_mask = perm_mask.permute(1, 2, 0) if perm_mask is not None else None
    target_mapping = target_mapping.permute(1, 2, 0) if target_mapping is not None else None

    mlen = mems[0].shape[0] if mems is not None and mems[0] is not None else 0
    klen = mlen + qlen

    dtype_float = self.dtype

    # Attention mask
    # causal attention mask
    if self.attn_type == "uni":
        attn_mask = self.create_mask(qlen, mlen)
        attn_mask = attn_mask[:, :, None, None]
    elif self.attn_type == "bi":
        attn_mask = None
    else:
        raise ValueError(f"Unsupported attention type: {self.attn_type}")

    # data mask: input mask & perm mask
    assert input_mask is None or attention_mask is None, "You can only use one of input_mask (uses 1 for padding) "
    "or attention_mask (uses 0 for padding, added for compatibility with BERT). Please choose one."
    if input_mask is None and attention_mask is not None:
        input_mask = 1.0 - attention_mask
    if input_mask is not None and perm_mask is not None:
        data_mask = input_mask[None] + perm_mask
    elif input_mask is not None and perm_mask is None:
        data_mask = input_mask[None]
    elif input_mask is None and perm_mask is not None:
        data_mask = perm_mask
    else:
        data_mask = None

    if data_mask is not None:
        # all mems can be attended to
        if mlen > 0:
            mems_mask = ops.zeros((data_mask.shape[0], mlen, bsz)).to(dtype=data_mask.dtype)
            data_mask = ops.cat([mems_mask, data_mask], axis=1)
        if attn_mask is None:
            attn_mask = data_mask[:, :, :, None]
        else:
            attn_mask += data_mask[:, :, :, None]

    if attn_mask is not None:
        attn_mask = (attn_mask > 0).to(dtype_float)

    if attn_mask is not None:
        non_tgt_mask = -ops.eye(qlen).to(dtype=attn_mask.dtype)
        if mlen > 0:
            non_tgt_mask = ops.cat([ops.zeros((qlen, mlen)).to(dtype=attn_mask.dtype), non_tgt_mask],
                                             axis=-1)
        non_tgt_mask = ((attn_mask + non_tgt_mask[:, :, None, None]) > 0).to(dtype=attn_mask.dtype)
    else:
        non_tgt_mask = None

    # Word embeddings and prepare h & g hidden states
    if inputs_embeds is not None:
        word_emb_k = inputs_embeds
    else:
        word_emb_k = self.word_embedding(input_ids)
    output_h = self.dropout(word_emb_k)
    if target_mapping is not None:
        word_emb_q = self.mask_emb.expand(target_mapping.shape[0], bsz, -1)
        # else:  # We removed the inp_q input which was same as target mapping
        #     inp_q_ext = inp_q[:, :, None]
        #     word_emb_q = inp_q_ext * self.mask_emb + (1 - inp_q_ext) * word_emb_k
        output_g = self.dropout(word_emb_q)
    else:
        output_g = None

    # Segment embedding
    if token_type_ids is not None:
        # Convert `token_type_ids` to one-hot `seg_mat`
        if mlen > 0:
            mem_pad = ops.zeros((mlen, bsz), dtype=mindspore.int64)
            cat_ids = ops.cat([mem_pad, token_type_ids], axis=0)
        else:
            cat_ids = token_type_ids

        # `1` indicates not in the same segment [qlen x klen x bsz]
        seg_mat = (token_type_ids[:, None] != cat_ids[None, :]).long()
        seg_mat = ops.one_hot(seg_mat, depth=2).to(dtype_float)
    else:
        seg_mat = None

    # Positional encoding
    pos_emb = self.relative_positional_encoding(qlen, klen, bsz=bsz)
    pos_emb = self.dropout(pos_emb)

    # Prepare head mask if needed
    # 1.0 in head_mask indicate we keep the head
    # attention_probs has shape bsz x n_heads x N x N
    # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads] (a head_mask for each layer)
    # and head_mask is converted to shape [num_hidden_layers x qlen x klen x bsz x n_head]
    if head_mask is not None:
        if head_mask.dim() == 1:
            head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(0).unsqueeze(0)
            head_mask = head_mask.expand(self.n_layer, -1, -1, -1, -1)
        elif head_mask.dim() == 2:
            head_mask = head_mask.unsqueeze(1).unsqueeze(1).unsqueeze(1)
        head_mask_dtype = next(iter(self.parameters_dict().items()))[1].dtype
        head_mask = head_mask.to(dtype=head_mask_dtype)
    else:
        head_mask = [None] * self.n_layer

    new_mems = ()
    if mems is None:
        mems = [None] * len(self.layer)

    attentions = [] if output_attentions else None
    hidden_states = [] if output_hidden_states else None
    for i, layer_module in enumerate(self.layer):
        if use_mems:
            # cache new mems
            new_mems = new_mems + (self.cache_mem(output_h, mems[i]),)
        if output_hidden_states:
            hidden_states.append((output_h, output_g) if output_g is not None else output_h)

        outputs = layer_module(
            output_h,
            output_g,
            attn_mask_h=non_tgt_mask,
            attn_mask_g=attn_mask,
            r=pos_emb,
            seg_mat=seg_mat,
            mems=mems[i],
            target_mapping=target_mapping,
            head_mask=head_mask[i],
            output_attentions=output_attentions,
        )
        output_h, output_g = outputs[:2]
        if output_attentions:
            attentions.append(outputs[2])

    # Add last hidden state
    if output_hidden_states:
        hidden_states.append((output_h, output_g) if output_g is not None else output_h)

    output = self.dropout(output_g if output_g is not None else output_h)

    # Prepare outputs, we swapaxes back here to shape [bsz, len, hidden_dim] (cf. beginning of forward() method)
    output = output.permute(1, 0, 2)

    if not use_mems:
        new_mems = None

    if output_hidden_states:
        if output_g is not None:
            hidden_states = tuple(h.permute(1, 0, 2) for hs in hidden_states for h in hs)
        else:
            hidden_states = tuple(hs.permute(1, 0, 2) for hs in hidden_states)

    if output_attentions:
        if target_mapping is not None:
            # when target_mapping is provided, there are 2-tuple of attentions
            attentions = tuple(
                tuple(att_stream.permute(2, 3, 0, 1) for att_stream in t) for t in attentions
            )
        else:
            attentions = tuple(t.permute(2, 3, 0, 1) for t in attentions)

    if not return_dict:
        return tuple(v for v in [output, new_mems, hidden_states, attentions] if v is not None)

    return XLNetModelOutput(
        last_hidden_state=output, mems=new_mems, hidden_states=hidden_states, attentions=attentions
    )

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.create_mask(qlen, mlen)

Creates causal attention mask. Float mask where 1.0 indicates masked, 0.0 indicates not-masked.

PARAMETER DESCRIPTION
qlen

Sequence length
mlen

Mask length

::

      same_length=False: same_length=True: <mlen > < qlen > <mlen > < qlen >
   ^ [0 0 0 0 0 1 1 1 1] [0 0 0 0 0 1 1 1 1]
     [0 0 0 0 0 0 1 1 1] [1 0 0 0 0 0 1 1 1]
qlen [0 0 0 0 0 0 0 1 1] [1 1 0 0 0 0 0 1 1]
     [0 0 0 0 0 0 0 0 1] [1 1 1 0 0 0 0 0 1]
   v [0 0 0 0 0 0 0 0 0] [1 1 1 1 0 0 0 0 0]
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
def create_mask(self, qlen, mlen):
    """
    Creates causal attention mask. Float mask where 1.0 indicates masked, 0.0 indicates not-masked.

    Args:
        qlen: Sequence length
        mlen: Mask length

    ::

              same_length=False: same_length=True: <mlen > < qlen > <mlen > < qlen >
           ^ [0 0 0 0 0 1 1 1 1] [0 0 0 0 0 1 1 1 1]
             [0 0 0 0 0 0 1 1 1] [1 0 0 0 0 0 1 1 1]
        qlen [0 0 0 0 0 0 0 1 1] [1 1 0 0 0 0 0 1 1]
             [0 0 0 0 0 0 0 0 1] [1 1 1 0 0 0 0 0 1]
           v [0 0 0 0 0 0 0 0 0] [1 1 1 1 0 0 0 0 0]

    """
    mask = ops.ones((qlen, qlen + mlen))
    if self.same_length:
        mask_lo = mask[:, :qlen].tril(-1)
        mask.triu_(mlen + 1)
        mask[:, :qlen] += mask_lo
    else:
        mask.triu_(mlen + 1)

    return mask

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.get_input_embeddings()

This method retrieves the input embeddings from the XLNetModel.

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class. The self parameter is required to access the word_embedding attribute.

TYPE: XLNetModel

RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
def get_input_embeddings(self):
    """
    This method retrieves the input embeddings from the XLNetModel.

    Args:
        self (XLNetModel): The instance of the XLNetModel class.
            The self parameter is required to access the word_embedding attribute.

    Returns:
        None.

    Raises:
        None.
    """
    return self.word_embedding

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.positional_embedding(pos_seq, inv_freq, bsz=None) staticmethod

This method is a static method in the class 'XLNetModel' and is used to generate positional embeddings for input sequences.

PARAMETER DESCRIPTION
pos_seq

A tensor containing the positional sequence.

  • Type: torch.Tensor
  • Purpose: This tensor represents the positions of the input tokens in the sequence.
  • Restrictions: None

TYPE: Tensor

inv_freq

A tensor containing the inverse frequency values.

  • Type: torch.Tensor
  • Purpose: This tensor represents the inverse frequency values to be used in the positional embedding calculation.
  • Restrictions: None

TYPE: Tensor

bsz

An optional parameter representing the batch size.

  • Type: int
  • Purpose: This parameter is used to expand the positional embeddings tensor if provided.
  • Restrictions: None

TYPE: int DEFAULT: None

RETURNS DESCRIPTION

torch.Tensor: A tensor containing the positional embeddings.

  • Type: torch.Tensor
  • Purpose: This tensor represents the positional embeddings for the input sequence.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
@staticmethod
def positional_embedding(pos_seq, inv_freq, bsz=None):
    """
    This method is a static method in the class 'XLNetModel' and is used to generate positional embeddings for input sequences.

    Args:
        pos_seq (torch.Tensor):
            A tensor containing the positional sequence.

            - Type: torch.Tensor
            - Purpose: This tensor represents the positions of the input tokens in the sequence.
            - Restrictions: None

        inv_freq (torch.Tensor):
            A tensor containing the inverse frequency values.

            - Type: torch.Tensor
            - Purpose: This tensor represents the inverse frequency values to be used in the positional
            embedding calculation.
            - Restrictions: None

        bsz (int, optional):
            An optional parameter representing the batch size.

            - Type: int
            - Purpose: This parameter is used to expand the positional embeddings tensor if provided.
            - Restrictions: None

    Returns:
        torch.Tensor:
            A tensor containing the positional embeddings.

            - Type: torch.Tensor
            - Purpose: This tensor represents the positional embeddings for the input sequence.

    Raises:
        None
    """
    sinusoid_inp = ops.einsum("i,d->id", pos_seq, inv_freq)
    pos_emb = ops.cat([ops.sin(sinusoid_inp), ops.cos(sinusoid_inp)], axis=-1)
    pos_emb = pos_emb[:, None, :]

    if bsz is not None:
        pos_emb = pos_emb.expand(-1, bsz, -1)

    return pos_emb

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.relative_positional_encoding(qlen, klen, bsz=None)

Encodes relative positional information for the XLNetModel.

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class.

TYPE: XLNetModel

qlen

The length of the query sequence.

TYPE: int

klen

The length of the key sequence.

TYPE: int

bsz

The batch size. Defaults to None.

TYPE: int DEFAULT: None

RETURNS DESCRIPTION

None
RAISES DESCRIPTION
ValueError

If the attn_type is not 'bi' or 'uni'.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
def relative_positional_encoding(self, qlen, klen, bsz=None):
    """
    Encodes relative positional information for the XLNetModel.

    Args:
        self (XLNetModel): The instance of the XLNetModel class.
        qlen (int): The length of the query sequence.
        klen (int): The length of the key sequence.
        bsz (int, optional): The batch size. Defaults to None.

    Returns:
        None

    Raises:
        ValueError: If the `attn_type` is not 'bi' or 'uni'.

    """
    # create relative positional encoding.
    freq_seq = ops.arange(0, self.d_model, 2.0, dtype=mindspore.int64).float()
    inv_freq = 1 / ops.pow(10000, (freq_seq / self.d_model))

    if self.attn_type == "bi":
        # beg, end = klen - 1, -qlen
        beg, end = klen, -qlen
    elif self.attn_type == "uni":
        # beg, end = klen - 1, -1
        beg, end = klen, -1
    else:
        raise ValueError(f"Unknown `attn_type` {self.attn_type}.")

    if self.bi_data:
        fwd_pos_seq = ops.arange(beg, end, -1.0, dtype=mindspore.int64).float()
        bwd_pos_seq = ops.arange(-beg, -end, 1.0, dtype=mindspore.int64).float()

        if self.clamp_len > 0:
            fwd_pos_seq = fwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)
            bwd_pos_seq = bwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)

        if bsz is not None:
            fwd_pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq, bsz // 2)
            bwd_pos_emb = self.positional_embedding(bwd_pos_seq, inv_freq, bsz // 2)
        else:
            fwd_pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq)
            bwd_pos_emb = self.positional_embedding(bwd_pos_seq, inv_freq)

        pos_emb = ops.cat([fwd_pos_emb, bwd_pos_emb], axis=1)
    else:
        fwd_pos_seq = ops.arange(beg, end, -1.0, dtype=mindspore.int64).float()
        if self.clamp_len > 0:
            fwd_pos_seq = fwd_pos_seq.clamp(-self.clamp_len, self.clamp_len)
        pos_emb = self.positional_embedding(fwd_pos_seq, inv_freq, bsz)

    return pos_emb

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModel.set_input_embeddings(new_embeddings)

Method to set new input embeddings for the XLNetModel.

PARAMETER DESCRIPTION
self

The instance of the XLNetModel class. This parameter is a reference to the current XLNetModel instance where the embeddings will be set.

TYPE: XLNetModel

new_embeddings

The new input embeddings to be assigned to the XLNetModel. This parameter represents the new embeddings that will replace the existing word embeddings in the model.

TYPE: any

RETURNS DESCRIPTION

None.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
def set_input_embeddings(self, new_embeddings):
    """
    Method to set new input embeddings for the XLNetModel.

    Args:
        self (XLNetModel): The instance of the XLNetModel class.
            This parameter is a reference to the current XLNetModel instance where the embeddings will be set.
        new_embeddings (any): The new input embeddings to be assigned to the XLNetModel.
            This parameter represents the new embeddings that will replace the existing word embeddings in the model.

    Returns:
        None.

    Raises:
        None.
    """
    self.word_embedding = new_embeddings

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetModelOutput dataclass

Bases: ModelOutput

Output type of [XLNetModel].

PARAMETER DESCRIPTION
last_hidden_state

Sequence of hidden-states at the last layer of the model.

num_predict corresponds to target_mapping.shape[1]. If target_mapping is None, then num_predict corresponds to sequence_length.

TYPE: `torch.FloatTensor` of shape `(batch_size, num_predict, hidden_size)`

mems

Contains pre-computed hidden-states. Can be used (see mems input) to speed up sequential decoding. The token ids which have their past given to this model should not be passed as input_ids as they have already been computed.

TYPE: `List[torch.FloatTensor]` of length `config.n_layers` DEFAULT: None

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
@dataclass
class XLNetModelOutput(ModelOutput):
    """
    Output type of [`XLNetModel`].

    Args:
        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, num_predict, hidden_size)`):
            Sequence of hidden-states at the last layer of the model.

            `num_predict` corresponds to `target_mapping.shape[1]`. If `target_mapping` is `None`, then `num_predict`
            corresponds to `sequence_length`.
        mems (`List[torch.FloatTensor]` of length `config.n_layers`):
            Contains pre-computed hidden-states. Can be used (see `mems` input) to speed up sequential decoding. The
            token ids which have their past given to this model should not be passed as `input_ids` as they have
            already been computed.
        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.
    """
    last_hidden_state: mindspore.Tensor
    mems: Optional[List[mindspore.Tensor]] = None
    hidden_states: Optional[Tuple[mindspore.Tensor, ...]] = None
    attentions: Optional[Tuple[mindspore.Tensor, ...]] = None

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetPreTrainedModel

Bases: PreTrainedModel

An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
class XLNetPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """
    config_class = XLNetConfig
    base_model_prefix = "transformer"

    def _init_weights(self, cell):
        """Initialize the weights."""
        if isinstance(cell, nn.Dense):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            cell.weight.set_data(initializer(Normal(self.config.initializer_range),
                                             cell.weight.shape, cell.weight.dtype))
            if cell.has_bias:
                cell.bias.set_data(initializer('zeros', cell.bias.shape, cell.bias.dtype))
        elif isinstance(cell, nn.Embedding):
            weight = np.random.normal(0.0, self.config.initializer_range, cell.weight.shape)
            if cell.padding_idx:
                weight[cell.padding_idx] = 0
            cell.weight.set_data(Tensor(weight, cell.weight.dtype))
        elif isinstance(cell, nn.LayerNorm):
            cell.weight.set_data(initializer('ones', cell.weight.shape, cell.weight.dtype))
            cell.bias.set_data(initializer('zeros', cell.bias.shape, cell.bias.dtype))
        elif isinstance(cell, XLNetRelativeAttention):
            for param in [
                cell.q,
                cell.k,
                cell.v,
                cell.o,
                cell.r,
                cell.r_r_bias,
                cell.r_s_bias,
                cell.r_w_bias,
                cell.seg_embed,
            ]:
                param.set_data(initializer(Normal(self.config.initializer_range),
                                           param.shape, param.dtype))
        elif isinstance(cell, XLNetModel):
            cell.mask_emb.set_data(initializer(Normal(self.config.initializer_range),
                                               cell.mask_emb.shape, cell.mask_emb.dtype))

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention

Bases: Cell

This class represents the relative attention mechanism used in XLNet model for sequence processing tasks.

The XLNetRelativeAttention class implements the core operations for performing relative positional attention in the XLNet model. It includes methods for initializing the attention mechanism, pruning attention heads, shifting for relative attention score calculation, and processing post-attention outputs.

ATTRIBUTE DESCRIPTION
n_head

Number of attention heads.

TYPE: int

d_head

Dimensionality of each attention head.

TYPE: int

d_model

Dimensionality of the model.

TYPE: int

scale

Scaling factor for attention scores.

TYPE: float

q

Query matrix for attention computation.

TYPE: Parameter

k

Key matrix for attention computation.

TYPE: Parameter

v

Value matrix for attention computation.

TYPE: Parameter

o

Output matrix for attention computation.

TYPE: Parameter

r

Relative position matrix.

TYPE: Parameter

r_r_bias

Relative position bias for rows.

TYPE: Parameter

r_s_bias

Relative position bias for segments.

TYPE: Parameter

r_w_bias

Relative position bias for columns.

TYPE: Parameter

seg_embed

Segment embedding matrix.

TYPE: Parameter

layer_norm

Layer normalization for model outputs.

TYPE: LayerNorm

dropout

Dropout layer for regularization.

TYPE: Dropout

METHOD DESCRIPTION
prune_heads

Method to prune specific attention heads (NotImplementedError).
rel_shift

Static method to perform relative shift for attention score calculation.
rel_shift_bnij

Static method to perform relative shift for attention score calculation with different axis.
rel_attn_core

Method for core relative positional attention operations.
post_attention

Method for post-attention processing.
construct

Method for constructing the attention mechanism with optional outputs.
Note

This class inherits from nn.Cell, which is a base class for neural network cells in the MindSpore framework.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
class XLNetRelativeAttention(nn.Cell):

    """This class represents the relative attention mechanism used in XLNet model for sequence processing tasks.

    The XLNetRelativeAttention class implements the core operations for performing relative positional attention
    in the XLNet model. It includes methods for initializing the attention mechanism, pruning attention heads,
    shifting for relative attention score calculation, and processing post-attention outputs.

    Attributes:
        n_head (int): Number of attention heads.
        d_head (int): Dimensionality of each attention head.
        d_model (int): Dimensionality of the model.
        scale (float): Scaling factor for attention scores.
        q (Parameter): Query matrix for attention computation.
        k (Parameter): Key matrix for attention computation.
        v (Parameter): Value matrix for attention computation.
        o (Parameter): Output matrix for attention computation.
        r (Parameter): Relative position matrix.
        r_r_bias (Parameter): Relative position bias for rows.
        r_s_bias (Parameter): Relative position bias for segments.
        r_w_bias (Parameter): Relative position bias for columns.
        seg_embed (Parameter): Segment embedding matrix.
        layer_norm (LayerNorm): Layer normalization for model outputs.
        dropout (Dropout): Dropout layer for regularization.

    Methods:
        prune_heads: Method to prune specific attention heads (NotImplementedError).
        rel_shift: Static method to perform relative shift for attention score calculation.
        rel_shift_bnij: Static method to perform relative shift for attention score calculation with
            different axis.
        rel_attn_core: Method for core relative positional attention operations.
        post_attention: Method for post-attention processing.
        construct: Method for constructing the attention mechanism with optional outputs.

    Note:
        This class inherits from nn.Cell, which is a base class for neural network cells in the MindSpore framework.
    """
    def __init__(self, config):
        '''
        Initializes the XLNetRelativeAttention class.

        Args:
            self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
            config: An object containing configuration parameters for the XLNetRelativeAttention model.
                It should have the following attributes:

                - d_model (int): The hidden size of the model.
                - n_head (int): The number of attention heads.
                - d_head (int): The size of each attention head.
                - layer_norm_eps (float): The epsilon value for layer normalization.
                - dropout (float): The dropout rate.

        Returns:
            None.

        Raises:
            ValueError: If the hidden size (config.d_model) is not a multiple of the number of attention heads
                (config.n_head).
        '''
        super().__init__()

        if config.d_model % config.n_head != 0:
            raise ValueError(
                f"The hidden size ({config.d_model}) is not a multiple of the number of attention "
                f"heads ({config.n_head}"
            )

        self.n_head = config.n_head
        self.d_head = config.d_head
        self.d_model = config.d_model
        self.scale = 1 / (config.d_head ** 0.5)

        self.q = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
        self.k = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
        self.v = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
        self.o = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
        self.r = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))

        self.r_r_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
        self.r_s_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
        self.r_w_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
        self.seg_embed = Parameter(ops.zeros((2, self.n_head, self.d_head), dtype=mindspore.float32))

        self.layer_norm = nn.LayerNorm(config.d_model, epsilon=config.layer_norm_eps)
        self.dropout = nn.Dropout(p=config.dropout)

    def prune_heads(self, heads):
        """
        This method prunes the given heads in the XLNetRelativeAttention class.

        Args:
            self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
            heads (int): The number of heads to be pruned from the attention mechanism. It should be a positive integer.

        Returns:
            None.

        Raises:
            NotImplementedError: This exception is raised if the method is called directly without being implemented
                in a subclass.
        """
        raise NotImplementedError

    @staticmethod
    def rel_shift(x, klen=-1):
        """perform relative shift to form the relative attention score."""
        x_size = x.shape

        x = x.reshape(x_size[1], x_size[0], x_size[2], x_size[3])
        x = x[1:, ...]
        x = x.reshape(x_size[0], x_size[1] - 1, x_size[2], x_size[3])
        # x = x[:, 0:klen, :, :]
        x = ops.index_select(x, 1, ops.arange(klen, dtype=mindspore.int64))

        return x

    @staticmethod
    def rel_shift_bnij(x, klen=-1):
        """
        Applies a relative shift to the input tensor along the batch and head dimensions in XLNetRelativeAttention class.

        Args:
            x (Tensor): Input tensor of shape (batch_size, seq_length, head_size, hidden_size).
            klen (int, optional): Length of the relative attention. Defaults to -1.

        Returns:
            None

        Raises:
            TypeError: If x is not a Tensor.
            ValueError: If x has an invalid shape or klen is not within the valid range.

        Note:
            - This method applies a relative shift operation to the input tensor `x`, rearranging the elements along
            the batch and head dimensions.
            - The input tensor `x` should have shape (batch_size, seq_length, head_size, hidden_size).
            - The relative shift is performed by reshaping the tensor, selecting specific indices, and reshaping
            it back to the original shape.
            - The relative shift helps in capturing the relative positions of tokens in the attention mechanism.
            - If `klen` is not provided, the default value of -1 is used which indicates that the relative attention
            length is not restricted.

        Example:
            ```python
            >>> x = Tensor([[1, 2, 3], [4, 5, 6]])
            >>> klen = 2
            >>> rel_shift_bnij(x, klen)
            ```
        """
        x_size = x.shape

        x = x.reshape(x_size[0], x_size[1], x_size[3], x_size[2])
        x = x[:, :, 1:, :]
        x = x.reshape(x_size[0], x_size[1], x_size[2], x_size[3] - 1)
        # Note: the tensor-slice form was faster in my testing than torch.index_select
        #       However, tracing doesn't like the nature of the slice, and if klen changes
        #       during the run then it'll fail, whereas index_select will be fine.
        x = ops.index_select(x, 3, ops.arange(klen, dtype=mindspore.int64))
        # x = x[:, :, :, :klen]

        return x

    def rel_attn_core(
            self,
            q_head,
            k_head_h,
            v_head_h,
            k_head_r,
            seg_mat=None,
            attn_mask=None,
            head_mask=None,
            output_attentions=False,
    ):
        """Core relative positional attention operations."""
        # content based attention score
        ac = ops.einsum("ibnd,jbnd->bnij", q_head + self.r_w_bias, k_head_h)

        # position based attention score
        bd = ops.einsum("ibnd,jbnd->bnij", q_head + self.r_r_bias, k_head_r)
        bd = self.rel_shift_bnij(bd, klen=ac.shape[3])

        # segment based attention score
        if seg_mat is None:
            ef = 0
        else:
            ef = ops.einsum("ibnd,snd->ibns", q_head + self.r_s_bias, self.seg_embed)
            ef = ops.einsum("ijbs,ibns->bnij", seg_mat, ef)

        # merge attention scores and perform masking
        attn_score = (ac + bd + ef) * self.scale
        if attn_mask is not None:
            # attn_score = attn_score * (1 - attn_mask) - 1e30 * attn_mask
            if attn_mask.dtype == mindspore.float16:
                attn_score = attn_score - 65500 * ops.einsum("ijbn->bnij", attn_mask)
            else:
                attn_score = attn_score - 1e30 * ops.einsum("ijbn->bnij", attn_mask)

        # attention probability
        attn_prob = ops.softmax(attn_score, axis=3)
        attn_prob = self.dropout(attn_prob)

        # Mask heads if we want to
        if head_mask is not None:
            attn_prob = attn_prob * ops.einsum("ijbn->bnij", head_mask)

        # attention output
        attn_vec = ops.einsum("bnij,jbnd->ibnd", attn_prob, v_head_h)

        if output_attentions:
            return attn_vec, ops.einsum("bnij->ijbn", attn_prob)

        return attn_vec

    def post_attention(self, h, attn_vec, residual=True):
        """Post-attention processing."""
        # post-attention projection (back to `d_model`)
        attn_out = ops.einsum("ibnd,hnd->ibh", attn_vec, self.o)

        attn_out = self.dropout(attn_out)
        if residual:
            attn_out = attn_out + h
        output = self.layer_norm(attn_out)

        return output

    def construct(
            self,
            h,
            g,
            attn_mask_h,
            attn_mask_g,
            r,
            seg_mat,
            mems=None,
            target_mapping=None,
            head_mask=None,
            output_attentions=False,
    ):
        """
        Constructs the attention vectors for the XLNetRelativeAttention module.

        Args:
            self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
            h (Tensor): The input tensor h of shape (batch_size, seq_length, hidden_size) representing the hidden states.
            g (Tensor): The input tensor g of shape (batch_size, seq_length, hidden_size) representing the hidden states
                of the global context.
            attn_mask_h (Tensor): The attention mask for h of shape (batch_size, seq_length, seq_length).
            attn_mask_g (Tensor): The attention mask for g of shape (batch_size, seq_length, seq_length).
            r (Tensor): The input tensor r of shape (batch_size, seq_length, hidden_size) representing the hidden states
                of the relative positions.
            seg_mat (Tensor): The input tensor seg_mat of shape (batch_size, seq_length, seq_length) representing the
                segment matrix.
            mems (Tensor, optional): The input tensor mems of shape (mem_len, batch_size, hidden_size) representing
                the memory states. Default is None.
            target_mapping (Tensor, optional): The input tensor target_mapping of shape (batch_size, seq_length, mem_len)
                representing the target mapping. Default is None.
            head_mask (Tensor, optional): The input tensor head_mask of shape (num_heads,) representing the mask for
                the attention heads. Default is None.
            output_attentions (bool, optional): Whether to output the attention probabilities. Default is False.

        Returns:
            Tuple:
                A tuple of two tensors (output_h, output_g) representing the output hidden states for h and g
                respectively. If output_attentions is True, the tuple also contains a tensor attn_prob of shape
                (num_heads, batch_size, seq_length, seq_length) representing the attention probabilities.

        Raises:
            None.
        """
        if g is not None:
            # Two-stream attention with relative positional encoding.
            # content based attention score
            if mems is not None and mems.dim() > 1:
                cat = ops.cat([mems, h], axis=0)
            else:
                cat = h

            # content-based key head
            k_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.k)

            # content-based value head
            v_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.v)

            # position-based key head
            k_head_r = ops.einsum("ibh,hnd->ibnd", r, self.r)

            # h-stream
            # content-stream query head
            q_head_h = ops.einsum("ibh,hnd->ibnd", h, self.q)

            # core attention ops
            attn_vec_h = self.rel_attn_core(
                q_head_h,
                k_head_h,
                v_head_h,
                k_head_r,
                seg_mat=seg_mat,
                attn_mask=attn_mask_h,
                head_mask=head_mask,
                output_attentions=output_attentions,
            )

            if output_attentions:
                attn_vec_h, attn_prob_h = attn_vec_h

            # post processing
            output_h = self.post_attention(h, attn_vec_h)

            # g-stream
            # query-stream query head
            q_head_g = ops.einsum("ibh,hnd->ibnd", g, self.q)

            # core attention ops
            if target_mapping is not None:
                q_head_g = ops.einsum("mbnd,mlb->lbnd", q_head_g, target_mapping)
                attn_vec_g = self.rel_attn_core(
                    q_head_g,
                    k_head_h,
                    v_head_h,
                    k_head_r,
                    seg_mat=seg_mat,
                    attn_mask=attn_mask_g,
                    head_mask=head_mask,
                    output_attentions=output_attentions,
                )

                if output_attentions:
                    attn_vec_g, attn_prob_g = attn_vec_g

                attn_vec_g = ops.einsum("lbnd,mlb->mbnd", attn_vec_g, target_mapping)
            else:
                attn_vec_g = self.rel_attn_core(
                    q_head_g,
                    k_head_h,
                    v_head_h,
                    k_head_r,
                    seg_mat=seg_mat,
                    attn_mask=attn_mask_g,
                    head_mask=head_mask,
                    output_attentions=output_attentions,
                )

                if output_attentions:
                    attn_vec_g, attn_prob_g = attn_vec_g

            # post processing
            output_g = self.post_attention(g, attn_vec_g)

            if output_attentions:
                attn_prob = attn_prob_h, attn_prob_g

        else:
            # Multi-head attention with relative positional encoding
            if mems is not None and mems.dim() > 1:
                cat = ops.cat([mems, h], axis=0)
            else:
                cat = h

            # content heads
            q_head_h = ops.einsum("ibh,hnd->ibnd", h, self.q)
            k_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.k)
            v_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.v)

            # positional heads
            # type casting for fp16 support
            k_head_r = ops.einsum("ibh,hnd->ibnd", r.type(self.r.dtype), self.r)

            # core attention ops
            attn_vec = self.rel_attn_core(
                q_head_h,
                k_head_h,
                v_head_h,
                k_head_r,
                seg_mat=seg_mat,
                attn_mask=attn_mask_h,
                head_mask=head_mask,
                output_attentions=output_attentions,
            )

            if output_attentions:
                attn_vec, attn_prob = attn_vec

            # post processing
            output_h = self.post_attention(h, attn_vec)
            output_g = None

        outputs = (output_h, output_g)
        if output_attentions:
            outputs = outputs + (attn_prob,)
        return outputs

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.__init__(config)

Initializes the XLNetRelativeAttention class.

PARAMETER DESCRIPTION
self

The instance of the XLNetRelativeAttention class.

TYPE: XLNetRelativeAttention

config

An object containing configuration parameters for the XLNetRelativeAttention model. It should have the following attributes:

  • d_model (int): The hidden size of the model.
  • n_head (int): The number of attention heads.
  • d_head (int): The size of each attention head.
  • layer_norm_eps (float): The epsilon value for layer normalization.
  • dropout (float): The dropout rate.
RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
ValueError

If the hidden size (config.d_model) is not a multiple of the number of attention heads (config.n_head).
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
def __init__(self, config):
    '''
    Initializes the XLNetRelativeAttention class.

    Args:
        self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
        config: An object containing configuration parameters for the XLNetRelativeAttention model.
            It should have the following attributes:

            - d_model (int): The hidden size of the model.
            - n_head (int): The number of attention heads.
            - d_head (int): The size of each attention head.
            - layer_norm_eps (float): The epsilon value for layer normalization.
            - dropout (float): The dropout rate.

    Returns:
        None.

    Raises:
        ValueError: If the hidden size (config.d_model) is not a multiple of the number of attention heads
            (config.n_head).
    '''
    super().__init__()

    if config.d_model % config.n_head != 0:
        raise ValueError(
            f"The hidden size ({config.d_model}) is not a multiple of the number of attention "
            f"heads ({config.n_head}"
        )

    self.n_head = config.n_head
    self.d_head = config.d_head
    self.d_model = config.d_model
    self.scale = 1 / (config.d_head ** 0.5)

    self.q = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
    self.k = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
    self.v = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
    self.o = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))
    self.r = Parameter(ops.zeros((config.d_model, self.n_head, self.d_head), dtype=mindspore.float32))

    self.r_r_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
    self.r_s_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
    self.r_w_bias = Parameter(ops.zeros((self.n_head, self.d_head), dtype=mindspore.float32))
    self.seg_embed = Parameter(ops.zeros((2, self.n_head, self.d_head), dtype=mindspore.float32))

    self.layer_norm = nn.LayerNorm(config.d_model, epsilon=config.layer_norm_eps)
    self.dropout = nn.Dropout(p=config.dropout)

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.construct(h, g, attn_mask_h, attn_mask_g, r, seg_mat, mems=None, target_mapping=None, head_mask=None, output_attentions=False)

Constructs the attention vectors for the XLNetRelativeAttention module.

PARAMETER DESCRIPTION
self

The instance of the XLNetRelativeAttention class.

TYPE: XLNetRelativeAttention

h

The input tensor h of shape (batch_size, seq_length, hidden_size) representing the hidden states.

TYPE: Tensor

g

The input tensor g of shape (batch_size, seq_length, hidden_size) representing the hidden states of the global context.

TYPE: Tensor

attn_mask_h

The attention mask for h of shape (batch_size, seq_length, seq_length).

TYPE: Tensor

attn_mask_g

The attention mask for g of shape (batch_size, seq_length, seq_length).

TYPE: Tensor

r

The input tensor r of shape (batch_size, seq_length, hidden_size) representing the hidden states of the relative positions.

TYPE: Tensor

seg_mat

The input tensor seg_mat of shape (batch_size, seq_length, seq_length) representing the segment matrix.

TYPE: Tensor

mems

The input tensor mems of shape (mem_len, batch_size, hidden_size) representing the memory states. Default is None.

TYPE: Tensor DEFAULT: None

target_mapping

The input tensor target_mapping of shape (batch_size, seq_length, mem_len) representing the target mapping. Default is None.

TYPE: Tensor DEFAULT: None

head_mask

The input tensor head_mask of shape (num_heads,) representing the mask for the attention heads. Default is None.

TYPE: Tensor DEFAULT: None

output_attentions

Whether to output the attention probabilities. Default is False.

TYPE: bool DEFAULT: False

RETURNS DESCRIPTION
Tuple

A tuple of two tensors (output_h, output_g) representing the output hidden states for h and g respectively. If output_attentions is True, the tuple also contains a tensor attn_prob of shape (num_heads, batch_size, seq_length, seq_length) representing the attention probabilities.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
def construct(
        self,
        h,
        g,
        attn_mask_h,
        attn_mask_g,
        r,
        seg_mat,
        mems=None,
        target_mapping=None,
        head_mask=None,
        output_attentions=False,
):
    """
    Constructs the attention vectors for the XLNetRelativeAttention module.

    Args:
        self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
        h (Tensor): The input tensor h of shape (batch_size, seq_length, hidden_size) representing the hidden states.
        g (Tensor): The input tensor g of shape (batch_size, seq_length, hidden_size) representing the hidden states
            of the global context.
        attn_mask_h (Tensor): The attention mask for h of shape (batch_size, seq_length, seq_length).
        attn_mask_g (Tensor): The attention mask for g of shape (batch_size, seq_length, seq_length).
        r (Tensor): The input tensor r of shape (batch_size, seq_length, hidden_size) representing the hidden states
            of the relative positions.
        seg_mat (Tensor): The input tensor seg_mat of shape (batch_size, seq_length, seq_length) representing the
            segment matrix.
        mems (Tensor, optional): The input tensor mems of shape (mem_len, batch_size, hidden_size) representing
            the memory states. Default is None.
        target_mapping (Tensor, optional): The input tensor target_mapping of shape (batch_size, seq_length, mem_len)
            representing the target mapping. Default is None.
        head_mask (Tensor, optional): The input tensor head_mask of shape (num_heads,) representing the mask for
            the attention heads. Default is None.
        output_attentions (bool, optional): Whether to output the attention probabilities. Default is False.

    Returns:
        Tuple:
            A tuple of two tensors (output_h, output_g) representing the output hidden states for h and g
            respectively. If output_attentions is True, the tuple also contains a tensor attn_prob of shape
            (num_heads, batch_size, seq_length, seq_length) representing the attention probabilities.

    Raises:
        None.
    """
    if g is not None:
        # Two-stream attention with relative positional encoding.
        # content based attention score
        if mems is not None and mems.dim() > 1:
            cat = ops.cat([mems, h], axis=0)
        else:
            cat = h

        # content-based key head
        k_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.k)

        # content-based value head
        v_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.v)

        # position-based key head
        k_head_r = ops.einsum("ibh,hnd->ibnd", r, self.r)

        # h-stream
        # content-stream query head
        q_head_h = ops.einsum("ibh,hnd->ibnd", h, self.q)

        # core attention ops
        attn_vec_h = self.rel_attn_core(
            q_head_h,
            k_head_h,
            v_head_h,
            k_head_r,
            seg_mat=seg_mat,
            attn_mask=attn_mask_h,
            head_mask=head_mask,
            output_attentions=output_attentions,
        )

        if output_attentions:
            attn_vec_h, attn_prob_h = attn_vec_h

        # post processing
        output_h = self.post_attention(h, attn_vec_h)

        # g-stream
        # query-stream query head
        q_head_g = ops.einsum("ibh,hnd->ibnd", g, self.q)

        # core attention ops
        if target_mapping is not None:
            q_head_g = ops.einsum("mbnd,mlb->lbnd", q_head_g, target_mapping)
            attn_vec_g = self.rel_attn_core(
                q_head_g,
                k_head_h,
                v_head_h,
                k_head_r,
                seg_mat=seg_mat,
                attn_mask=attn_mask_g,
                head_mask=head_mask,
                output_attentions=output_attentions,
            )

            if output_attentions:
                attn_vec_g, attn_prob_g = attn_vec_g

            attn_vec_g = ops.einsum("lbnd,mlb->mbnd", attn_vec_g, target_mapping)
        else:
            attn_vec_g = self.rel_attn_core(
                q_head_g,
                k_head_h,
                v_head_h,
                k_head_r,
                seg_mat=seg_mat,
                attn_mask=attn_mask_g,
                head_mask=head_mask,
                output_attentions=output_attentions,
            )

            if output_attentions:
                attn_vec_g, attn_prob_g = attn_vec_g

        # post processing
        output_g = self.post_attention(g, attn_vec_g)

        if output_attentions:
            attn_prob = attn_prob_h, attn_prob_g

    else:
        # Multi-head attention with relative positional encoding
        if mems is not None and mems.dim() > 1:
            cat = ops.cat([mems, h], axis=0)
        else:
            cat = h

        # content heads
        q_head_h = ops.einsum("ibh,hnd->ibnd", h, self.q)
        k_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.k)
        v_head_h = ops.einsum("ibh,hnd->ibnd", cat, self.v)

        # positional heads
        # type casting for fp16 support
        k_head_r = ops.einsum("ibh,hnd->ibnd", r.type(self.r.dtype), self.r)

        # core attention ops
        attn_vec = self.rel_attn_core(
            q_head_h,
            k_head_h,
            v_head_h,
            k_head_r,
            seg_mat=seg_mat,
            attn_mask=attn_mask_h,
            head_mask=head_mask,
            output_attentions=output_attentions,
        )

        if output_attentions:
            attn_vec, attn_prob = attn_vec

        # post processing
        output_h = self.post_attention(h, attn_vec)
        output_g = None

    outputs = (output_h, output_g)
    if output_attentions:
        outputs = outputs + (attn_prob,)
    return outputs

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.post_attention(h, attn_vec, residual=True)

Post-attention processing.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
261
262
263
264
265
266
267
268
269
270
271
def post_attention(self, h, attn_vec, residual=True):
    """Post-attention processing."""
    # post-attention projection (back to `d_model`)
    attn_out = ops.einsum("ibnd,hnd->ibh", attn_vec, self.o)

    attn_out = self.dropout(attn_out)
    if residual:
        attn_out = attn_out + h
    output = self.layer_norm(attn_out)

    return output

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.prune_heads(heads)

This method prunes the given heads in the XLNetRelativeAttention class.

PARAMETER DESCRIPTION
self

The instance of the XLNetRelativeAttention class.

TYPE: XLNetRelativeAttention

heads

The number of heads to be pruned from the attention mechanism. It should be a positive integer.

TYPE: int

RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
NotImplementedError

This exception is raised if the method is called directly without being implemented in a subclass.
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
def prune_heads(self, heads):
    """
    This method prunes the given heads in the XLNetRelativeAttention class.

    Args:
        self (XLNetRelativeAttention): The instance of the XLNetRelativeAttention class.
        heads (int): The number of heads to be pruned from the attention mechanism. It should be a positive integer.

    Returns:
        None.

    Raises:
        NotImplementedError: This exception is raised if the method is called directly without being implemented
            in a subclass.
    """
    raise NotImplementedError

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.rel_attn_core(q_head, k_head_h, v_head_h, k_head_r, seg_mat=None, attn_mask=None, head_mask=None, output_attentions=False)

Core relative positional attention operations.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
def rel_attn_core(
        self,
        q_head,
        k_head_h,
        v_head_h,
        k_head_r,
        seg_mat=None,
        attn_mask=None,
        head_mask=None,
        output_attentions=False,
):
    """Core relative positional attention operations."""
    # content based attention score
    ac = ops.einsum("ibnd,jbnd->bnij", q_head + self.r_w_bias, k_head_h)

    # position based attention score
    bd = ops.einsum("ibnd,jbnd->bnij", q_head + self.r_r_bias, k_head_r)
    bd = self.rel_shift_bnij(bd, klen=ac.shape[3])

    # segment based attention score
    if seg_mat is None:
        ef = 0
    else:
        ef = ops.einsum("ibnd,snd->ibns", q_head + self.r_s_bias, self.seg_embed)
        ef = ops.einsum("ijbs,ibns->bnij", seg_mat, ef)

    # merge attention scores and perform masking
    attn_score = (ac + bd + ef) * self.scale
    if attn_mask is not None:
        # attn_score = attn_score * (1 - attn_mask) - 1e30 * attn_mask
        if attn_mask.dtype == mindspore.float16:
            attn_score = attn_score - 65500 * ops.einsum("ijbn->bnij", attn_mask)
        else:
            attn_score = attn_score - 1e30 * ops.einsum("ijbn->bnij", attn_mask)

    # attention probability
    attn_prob = ops.softmax(attn_score, axis=3)
    attn_prob = self.dropout(attn_prob)

    # Mask heads if we want to
    if head_mask is not None:
        attn_prob = attn_prob * ops.einsum("ijbn->bnij", head_mask)

    # attention output
    attn_vec = ops.einsum("bnij,jbnd->ibnd", attn_prob, v_head_h)

    if output_attentions:
        return attn_vec, ops.einsum("bnij->ijbn", attn_prob)

    return attn_vec

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.rel_shift(x, klen=-1) staticmethod

perform relative shift to form the relative attention score.

Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
151
152
153
154
155
156
157
158
159
160
161
162
@staticmethod
def rel_shift(x, klen=-1):
    """perform relative shift to form the relative attention score."""
    x_size = x.shape

    x = x.reshape(x_size[1], x_size[0], x_size[2], x_size[3])
    x = x[1:, ...]
    x = x.reshape(x_size[0], x_size[1] - 1, x_size[2], x_size[3])
    # x = x[:, 0:klen, :, :]
    x = ops.index_select(x, 1, ops.arange(klen, dtype=mindspore.int64))

    return x

mindnlp.transformers.models.xlnet.modeling_xlnet.XLNetRelativeAttention.rel_shift_bnij(x, klen=-1) staticmethod

Applies a relative shift to the input tensor along the batch and head dimensions in XLNetRelativeAttention class.

PARAMETER DESCRIPTION
x

Input tensor of shape (batch_size, seq_length, head_size, hidden_size).

TYPE: Tensor

klen

Length of the relative attention. Defaults to -1.

TYPE: int DEFAULT: -1

RETURNS DESCRIPTION

None
RAISES DESCRIPTION
TypeError

If x is not a Tensor.
ValueError

If x has an invalid shape or klen is not within the valid range.
Note
  • This method applies a relative shift operation to the input tensor x, rearranging the elements along the batch and head dimensions.
  • The input tensor x should have shape (batch_size, seq_length, head_size, hidden_size).
  • The relative shift is performed by reshaping the tensor, selecting specific indices, and reshaping it back to the original shape.
  • The relative shift helps in capturing the relative positions of tokens in the attention mechanism.
  • If klen is not provided, the default value of -1 is used which indicates that the relative attention length is not restricted.
Example 
>>> x = Tensor([[1, 2, 3], [4, 5, 6]])
>>> klen = 2
>>> rel_shift_bnij(x, klen)
Source code in mindnlp/transformers/models/xlnet/modeling_xlnet.py 
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
@staticmethod
def rel_shift_bnij(x, klen=-1):
    """
    Applies a relative shift to the input tensor along the batch and head dimensions in XLNetRelativeAttention class.

    Args:
        x (Tensor): Input tensor of shape (batch_size, seq_length, head_size, hidden_size).
        klen (int, optional): Length of the relative attention. Defaults to -1.

    Returns:
        None

    Raises:
        TypeError: If x is not a Tensor.
        ValueError: If x has an invalid shape or klen is not within the valid range.

    Note:
        - This method applies a relative shift operation to the input tensor `x`, rearranging the elements along
        the batch and head dimensions.
        - The input tensor `x` should have shape (batch_size, seq_length, head_size, hidden_size).
        - The relative shift is performed by reshaping the tensor, selecting specific indices, and reshaping
        it back to the original shape.
        - The relative shift helps in capturing the relative positions of tokens in the attention mechanism.
        - If `klen` is not provided, the default value of -1 is used which indicates that the relative attention
        length is not restricted.

    Example:
        ```python
        >>> x = Tensor([[1, 2, 3], [4, 5, 6]])
        >>> klen = 2
        >>> rel_shift_bnij(x, klen)
        ```
    """
    x_size = x.shape

    x = x.reshape(x_size[0], x_size[1], x_size[3], x_size[2])
    x = x[:, :, 1:, :]
    x = x.reshape(x_size[0], x_size[1], x_size[2], x_size[3] - 1)
    # Note: the tensor-slice form was faster in my testing than torch.index_select
    #       However, tracing doesn't like the nature of the slice, and if klen changes
    #       during the run then it'll fail, whereas index_select will be fine.
    x = ops.index_select(x, 3, ops.arange(klen, dtype=mindspore.int64))
    # x = x[:, :, :, :klen]

    return x

mindnlp.transformers.models.xlnet.configuration_xlnet

XLNet configuration

mindnlp.transformers.models.xlnet.configuration_xlnet.XLNetConfig

Bases: PretrainedConfig

Configuration for XLNet

Source code in mindnlp/transformers/models/xlnet/configuration_xlnet.py 
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
class XLNetConfig(PretrainedConfig):
    """
    Configuration for XLNet
    """
    model_type = "xlnet"
    keys_to_ignore_at_inference = ["mems"]
    attribute_map = {
        "n_token": "vocab_size",  # Backward compatibility
        "hidden_size": "d_model",
        "num_attention_heads": "n_head",
        "num_hidden_layers": "n_layer",
    }

    def __init__(
            self,
            vocab_size=32000,
            d_model=1024,
            n_layer=24,
            n_head=16,
            d_inner=4096,
            ff_activation="gelu",
            untie_r=True,
            attn_type="bi",
            initializer_range=0.02,
            layer_norm_eps=1e-12,
            dropout=0.1,
            mem_len=512,
            reuse_len=None,
            use_mems_eval=True,
            use_mems_train=False,
            bi_data=False,
            clamp_len=-1,
            same_length=False,
            summary_type="last",
            summary_use_proj=True,
            summary_activation="tanh",
            summary_last_dropout=0.1,
            start_n_top=5,
            end_n_top=5,
            pad_token_id=5,
            bos_token_id=1,
            eos_token_id=2,
            **kwargs,
    ):
        """Constructs XLNetConfig."""
        self.vocab_size = vocab_size
        self.d_model = d_model
        self.n_layer = n_layer
        self.n_head = n_head
        if d_model % n_head != 0:
            raise ValueError(f"'d_model % n_head' ({d_model % n_head}) should be equal to 0")
        if "d_head" in kwargs:
            if kwargs["d_head"] != d_model // n_head:
                raise ValueError(
                    f"`d_head` ({kwargs['d_head']}) should be equal to `d_model // n_head` ({d_model // n_head})"
                )
        self.d_head = d_model // n_head
        self.ff_activation = ff_activation
        self.d_inner = d_inner
        self.untie_r = untie_r
        self.attn_type = attn_type

        self.initializer_range = initializer_range
        self.layer_norm_eps = layer_norm_eps

        self.dropout = dropout
        self.mem_len = mem_len
        self.reuse_len = reuse_len
        self.bi_data = bi_data
        self.clamp_len = clamp_len
        self.same_length = same_length

        self.summary_type = summary_type
        self.summary_use_proj = summary_use_proj
        self.summary_activation = summary_activation
        self.summary_last_dropout = summary_last_dropout
        self.start_n_top = start_n_top
        self.end_n_top = end_n_top

        self.bos_token_id = bos_token_id
        self.pad_token_id = pad_token_id
        self.eos_token_id = eos_token_id

        if "use_cache" in kwargs:
            warnings.warn(
                "The `use_cache` argument is deprecated and will be removed in a future version, use `use_mems_eval`"
                " instead.",
                FutureWarning,
            )
            use_mems_eval = kwargs["use_cache"]

        self.use_mems_eval = use_mems_eval
        self.use_mems_train = use_mems_train
        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

    @property
    def max_position_embeddings(self):
        """
        This method returns the maximum position embeddings for the XLNet model.

        Args:
            self (XLNetConfig): The instance of the XLNetConfig class.

        Returns:
            None: This method does not return any specific value, as it only logs a message and returns -1.

        Raises:
            None
        """
        logger.info(f"The model {self.model_type} is one of the few models that has no sequence length limit.")
        return -1

    @max_position_embeddings.setter
    def max_position_embeddings(self, value):
        """
        Sets the maximum position embeddings for the XLNetConfig class.

        Args:
            self (XLNetConfig): An instance of the XLNetConfig class.
            value: The desired value for the maximum position embeddings. It should be an integer.

        Returns:
            None.

        Raises:
            NotImplementedError: This exception is raised when trying to set the maximum position embeddings for
                the XLNetConfig class. Since the model type is one of the few models that has no sequence length
                limit, setting the maximum position embeddings is not allowed.

        Note:
            The model type should be specified before using this method.
        """
        # Message copied from Transformer-XL documentation
        raise NotImplementedError(
            f"The model {self.model_type} is one of the few models that has no sequence length limit."
        )

mindnlp.transformers.models.xlnet.configuration_xlnet.XLNetConfig.max_position_embeddings property writable

This method returns the maximum position embeddings for the XLNet model.

PARAMETER DESCRIPTION
self

The instance of the XLNetConfig class.

TYPE: XLNetConfig

RETURNS DESCRIPTION
None

This method does not return any specific value, as it only logs a message and returns -1.

mindnlp.transformers.models.xlnet.configuration_xlnet.XLNetConfig.__init__(vocab_size=32000, d_model=1024, n_layer=24, n_head=16, d_inner=4096, ff_activation='gelu', untie_r=True, attn_type='bi', initializer_range=0.02, layer_norm_eps=1e-12, dropout=0.1, mem_len=512, reuse_len=None, use_mems_eval=True, use_mems_train=False, bi_data=False, clamp_len=-1, same_length=False, summary_type='last', summary_use_proj=True, summary_activation='tanh', summary_last_dropout=0.1, start_n_top=5, end_n_top=5, pad_token_id=5, bos_token_id=1, eos_token_id=2, **kwargs)

Constructs XLNetConfig.

Source code in mindnlp/transformers/models/xlnet/configuration_xlnet.py 
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
def __init__(
        self,
        vocab_size=32000,
        d_model=1024,
        n_layer=24,
        n_head=16,
        d_inner=4096,
        ff_activation="gelu",
        untie_r=True,
        attn_type="bi",
        initializer_range=0.02,
        layer_norm_eps=1e-12,
        dropout=0.1,
        mem_len=512,
        reuse_len=None,
        use_mems_eval=True,
        use_mems_train=False,
        bi_data=False,
        clamp_len=-1,
        same_length=False,
        summary_type="last",
        summary_use_proj=True,
        summary_activation="tanh",
        summary_last_dropout=0.1,
        start_n_top=5,
        end_n_top=5,
        pad_token_id=5,
        bos_token_id=1,
        eos_token_id=2,
        **kwargs,
):
    """Constructs XLNetConfig."""
    self.vocab_size = vocab_size
    self.d_model = d_model
    self.n_layer = n_layer
    self.n_head = n_head
    if d_model % n_head != 0:
        raise ValueError(f"'d_model % n_head' ({d_model % n_head}) should be equal to 0")
    if "d_head" in kwargs:
        if kwargs["d_head"] != d_model // n_head:
            raise ValueError(
                f"`d_head` ({kwargs['d_head']}) should be equal to `d_model // n_head` ({d_model // n_head})"
            )
    self.d_head = d_model // n_head
    self.ff_activation = ff_activation
    self.d_inner = d_inner
    self.untie_r = untie_r
    self.attn_type = attn_type

    self.initializer_range = initializer_range
    self.layer_norm_eps = layer_norm_eps

    self.dropout = dropout
    self.mem_len = mem_len
    self.reuse_len = reuse_len
    self.bi_data = bi_data
    self.clamp_len = clamp_len
    self.same_length = same_length

    self.summary_type = summary_type
    self.summary_use_proj = summary_use_proj
    self.summary_activation = summary_activation
    self.summary_last_dropout = summary_last_dropout
    self.start_n_top = start_n_top
    self.end_n_top = end_n_top

    self.bos_token_id = bos_token_id
    self.pad_token_id = pad_token_id
    self.eos_token_id = eos_token_id

    if "use_cache" in kwargs:
        warnings.warn(
            "The `use_cache` argument is deprecated and will be removed in a future version, use `use_mems_eval`"
            " instead.",
            FutureWarning,
        )
        use_mems_eval = kwargs["use_cache"]

    self.use_mems_eval = use_mems_eval
    self.use_mems_train = use_mems_train
    super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

mindnlp.transformers.models.xlnet.tokenization_xlnet

Tokenization classes for XLNet model.

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer

Bases: PreTrainedTokenizer

Construct an XLNet tokenizer. Based on SentencePiece.

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

PARAMETER DESCRIPTION
vocab_file

SentencePiece file (generally has a .spm extension) that contains the vocabulary necessary to instantiate a tokenizer.

TYPE: `str`

do_lower_case

Whether to lowercase the input when tokenizing.

TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False

remove_space

Whether to strip the text when tokenizing (removing excess spaces before and after the string).

TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True

keep_accents

Whether to keep accents when tokenizing.

TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False

bos_token

The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.

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.

TYPE: `str`, *optional*, defaults to `"<s>"` DEFAULT: '<s>'

eos_token

The end of sequence token.

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

TYPE: `str`, *optional*, defaults to `"</s>"` DEFAULT: '</s>'

unk_token

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.

TYPE: `str`, *optional*, defaults to `"<unk>"` DEFAULT: '<unk>'

sep_token

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.

TYPE: `str`, *optional*, defaults to `"<sep>"` DEFAULT: '<sep>'

pad_token

The token used for padding, for example when batching sequences of different lengths.

TYPE: `str`, *optional*, defaults to `"<pad>"` DEFAULT: '<pad>'

cls_token

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.

TYPE: `str`, *optional*, defaults to `"<cls>"` DEFAULT: '<cls>'

mask_token

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.

TYPE: `str`, *optional*, defaults to `"<mask>"` DEFAULT: '<mask>'

additional_special_tokens

Additional special tokens used by the tokenizer.

TYPE: `List[str]`, *optional*, defaults to `['<eop>', '<eod>']` DEFAULT: ['<eop>', '<eod>']

sp_model_kwargs

Will be passed to the SentencePieceProcessor.__init__() method. The Python wrapper for SentencePiece can be used, among other things, to set:

  • enable_sampling: Enable subword regularization.

  • nbest_size: Sampling parameters for unigram. Invalid for BPE-Dropout.

    • nbest_size = {0,1}: No sampling is performed.
    • nbest_size > 1: samples from the nbest_size results.
    • nbest_size < 0: assuming that nbest_size is infinite and samples from the all hypothesis (lattice) using forward-filtering-and-backward-sampling algorithm.

  • alpha: Smoothing parameter for unigram sampling, and dropout probability of merge operations for BPE-dropout.

TYPE: `dict`, *optional* DEFAULT: None

ATTRIBUTE DESCRIPTION
sp_model

The SentencePiece processor that is used for every conversion (string, tokens and IDs).

TYPE: `SentencePieceProcessor`

Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
class XLNetTokenizer(PreTrainedTokenizer):
    """
    Construct an XLNet tokenizer. Based on [SentencePiece](https://github.com/google/sentencepiece).

    This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to
    this superclass for more information regarding those methods.

    Args:
        vocab_file (`str`):
            [SentencePiece](https://github.com/google/sentencepiece) file (generally has a .spm extension) that
            contains the vocabulary necessary to instantiate a tokenizer.
        do_lower_case (`bool`, *optional*, defaults to `False`):
            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.
        bos_token (`str`, *optional*, defaults to `"<s>"`):
            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.

            <Tip>

            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`.

            </Tip>

        eos_token (`str`, *optional*, defaults to `"</s>"`):
            The end of sequence token.

            <Tip>

            When building a sequence using special tokens, this is not the token that is used for the end of sequence.
            The token used is the `sep_token`.

            </Tip>

        unk_token (`str`, *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.
        sep_token (`str`, *optional*, defaults to `"<sep>"`):
            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 (`str`, *optional*, defaults to `"<pad>"`):
            The token used for padding, for example when batching sequences of different lengths.
        cls_token (`str`, *optional*, defaults to `"<cls>"`):
            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 (`str`, *optional*, defaults to `"<mask>"`):
            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 `['<eop>', '<eod>']`):
            Additional special tokens used by the tokenizer.
        sp_model_kwargs (`dict`, *optional*):
            Will be passed to the `SentencePieceProcessor.__init__()` method. The [Python wrapper for
            SentencePiece](https://github.com/google/sentencepiece/tree/master/python) can be used, among other things,
            to set:

            - `enable_sampling`: Enable subword regularization.
            - `nbest_size`: Sampling parameters for unigram. Invalid for BPE-Dropout.

                - `nbest_size = {0,1}`: No sampling is performed.
                - `nbest_size > 1`: samples from the nbest_size results.
                - `nbest_size < 0`: assuming that nbest_size is infinite and samples from the all hypothesis (lattice)
                using forward-filtering-and-backward-sampling algorithm.

            - `alpha`: Smoothing parameter for unigram sampling, and dropout probability of merge operations for
            BPE-dropout.

    Attributes:
        sp_model (`SentencePieceProcessor`):
            The *SentencePiece* processor that is used for every conversion (string, tokens and IDs).
    """
    vocab_files_names = VOCAB_FILES_NAMES
    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
    padding_side = "left"

    def __init__(
            self,
            vocab_file,
            do_lower_case=False,
            remove_space=True,
            keep_accents=False,
            bos_token="<s>",
            eos_token="</s>",
            unk_token="<unk>",
            sep_token="<sep>",
            pad_token="<pad>",
            cls_token="<cls>",
            mask_token="<mask>",
            additional_special_tokens=["<eop>", "<eod>"],
            sp_model_kwargs: Optional[Dict[str, Any]] = None,
            **kwargs,
    ) -> None:
        """
        Initialize an XLNetTokenizer object.

        Args:
            vocab_file (str): Path to the vocabulary file.
            do_lower_case (bool, optional): Whether to lowercase the input tokens. Defaults to False.
            remove_space (bool, optional): Whether to remove spaces in the input tokens. Defaults to True.
            keep_accents (bool, optional): Whether to keep accents in the input tokens. Defaults to False.
            bos_token (str, optional): Beginning of sentence token. Defaults to '<s>'.
            eos_token (str, optional): End of sentence token. Defaults to '</s>'.
            unk_token (str, optional): Unknown token. Defaults to '<unk>'.
            sep_token (str, optional): Separator token. Defaults to '<sep>'.
            pad_token (str, optional): Padding token. Defaults to '<pad>'.
            cls_token (str, optional): Classification token. Defaults to '<cls>'.
            mask_token (str, optional): Mask token. Defaults to '<mask>'.
            additional_special_tokens (list, optional): Additional special tokens to include.
                Defaults to ['<eop>', '<eod>'].
            sp_model_kwargs (Dict[str, Any], optional): SentencePiece model keyword arguments. Defaults to None.
            **kwargs: Additional keyword arguments.

        Returns:
            None

        Raises:
            TypeError: If the mask_token is not a string.
        """
        # Mask token behave like a normal word, i.e. include the space before it
        mask_token = AddedToken(mask_token, lstrip=True, special=True) if isinstance(mask_token, str) else mask_token

        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs

        self.do_lower_case = do_lower_case
        self.remove_space = remove_space
        self.keep_accents = keep_accents
        self.vocab_file = vocab_file

        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
        self.sp_model.Load(vocab_file)

        super().__init__(
            do_lower_case=do_lower_case,
            remove_space=remove_space,
            keep_accents=keep_accents,
            bos_token=bos_token,
            eos_token=eos_token,
            unk_token=unk_token,
            sep_token=sep_token,
            pad_token=pad_token,
            cls_token=cls_token,
            mask_token=mask_token,
            additional_special_tokens=additional_special_tokens,
            sp_model_kwargs=self.sp_model_kwargs,
            **kwargs,
        )

        self._pad_token_type_id = 3

    @property
    def vocab_size(self):
        """
        Returns the vocabulary size of the XLNetTokenizer.

        Args:
            self (XLNetTokenizer): An instance of the XLNetTokenizer class.

        Returns:
            int: The vocabulary size of the tokenizer.

        Raises:
            None

        This method calculates and returns the size of the vocabulary used by the XLNetTokenizer.
        The vocabulary size is determined by the number of unique tokens present in the tokenizer's sp_model.

        Note:
            The vocabulary size represents the number of distinct tokens that the tokenizer can recognize and encode.

        Example:
            ```python
            >>> tokenizer = XLNetTokenizer()
            >>> size = tokenizer.vocab_size()
            >>> print(size)
            32000
            ```
        """
        return len(self.sp_model)

    def get_vocab(self):
        """
        Returns the vocabulary of the XLNetTokenizer.

        Args:
            self: The instance of the XLNetTokenizer class.

        Returns:
            dict:
                A dictionary containing the vocabulary of the XLNetTokenizer.
                The keys of the dictionary are the tokens, and the values are their corresponding indices.

        Raises:
            None.

        Example:
            ```python
            >>> tokenizer = XLNetTokenizer()
            >>> tokenizer.get_vocab()
            {'<s>': 0, '<pad>': 1, '</s>': 2, '<unk>': 3, '<mask>': 4, 'hello': 5, 'world': 6}
            ```
        """
        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
        vocab.update(self.added_tokens_encoder)
        return vocab

    def __getstate__(self):
        """
        Method '__getstate__' in the class 'XLNetTokenizer'.

        Args:
            self (XLNetTokenizer): The instance of the XLNetTokenizer class.
                Represents the current XLNetTokenizer object.

        Returns:
            dict: This method returns a dictionary representing the state of the XLNetTokenizer object.
                The 'sp_model' key in the dictionary is set to None before returning.

        Raises:
            None.
        """
        state = self.__dict__.copy()
        state["sp_model"] = None
        return state

    def __setstate__(self, d):
        """
        This method __setstate__ is defined in the class XLNetTokenizer and is used to set the internal state of the
        object based on the provided dictionary 'd'.

        Args:
            self (XLNetTokenizer): The instance of the XLNetTokenizer class.
            d (dict): A dictionary containing the state information to be set. The keys and values in the dictionary
                are used to update the internal state of the XLNetTokenizer object.

        Returns:
            None. This method does not return any value.

        Raises:
            None:
                However, potential exceptions that could be raised include:

                - AttributeError: If the 'sp_model_kwargs' attribute is not found within the XLNetTokenizer object.
                - TypeError: If the provided 'd' parameter is not a dictionary.
                - Other exceptions related to the SentencePieceProcessor object creation or loading process may be
                raised from the spm.SentencePieceProcessor constructor or Load method.
        """
        self.__dict__ = d

        # for backward compatibility
        if not hasattr(self, "sp_model_kwargs"):
            self.sp_model_kwargs = {}

        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
        self.sp_model.Load(self.vocab_file)

    def preprocess_text(self, inputs):
        """
        This method preprocesses the input text according to the specified configuration settings.

        Args:
            self (XLNetTokenizer): The instance of the XLNetTokenizer class.
            inputs (str): The input text to be preprocessed. It should be a string representation.

        Returns:
            str: The preprocessed text based on the applied configuration settings.

        Raises:
            None
        """
        if self.remove_space:
            outputs = " ".join(inputs.strip().split())
        else:
            outputs = inputs
        outputs = outputs.replace("``", '"').replace("''", '"')

        if not self.keep_accents:
            outputs = unicodedata.normalize("NFKD", outputs)
            outputs = "".join([c for c in outputs if not unicodedata.combining(c)])
        if self.do_lower_case:
            outputs = outputs.lower()

        return outputs

    def _tokenize(self, text: str) -> List[str]:
        """Tokenize a string."""
        text = self.preprocess_text(text)
        pieces = self.sp_model.encode(text, out_type=str)
        new_pieces = []
        for piece in pieces:
            if len(piece) > 1 and piece[-1] == str(",") and piece[-2].isdigit():
                cur_pieces = self.sp_model.EncodeAsPieces(piece[:-1].replace(SPIECE_UNDERLINE, ""))
                if piece[0] != SPIECE_UNDERLINE and cur_pieces[0][0] == SPIECE_UNDERLINE:
                    if len(cur_pieces[0]) == 1:
                        cur_pieces = cur_pieces[1:]
                    else:
                        cur_pieces[0] = cur_pieces[0][1:]
                cur_pieces.append(piece[-1])
                new_pieces.extend(cur_pieces)
            else:
                new_pieces.append(piece)

        return new_pieces

    def _convert_token_to_id(self, token):
        """Converts a token (str) in an id using the vocab."""
        return self.sp_model.PieceToId(token)

    def _convert_id_to_token(self, index):
        """Converts an index (integer) in a token (str) using the vocab."""
        return self.sp_model.IdToPiece(index)

    def convert_tokens_to_string(self, tokens):
        """Converts a sequence of tokens (strings for sub-words) in a single string."""
        out_string = "".join(tokens).replace(SPIECE_UNDERLINE, " ").strip()
        return out_string

    def _decode(
            self,
            token_ids: List[int],
            skip_special_tokens: bool = False,
            clean_up_tokenization_spaces: bool = None,
            spaces_between_special_tokens: bool = True,
            **kwargs,
    ) -> str:
        """
        This method decodes a list of token IDs into a string representation.

        Args:
            self: The instance of the XLNetTokenizer class.
            token_ids (List[int]): A list of token IDs to be decoded into a string.
            skip_special_tokens (bool): A flag indicating whether to skip special tokens during decoding.
                Defaults to False.
            clean_up_tokenization_spaces (bool): A flag indicating whether to clean up tokenization spaces in the
                decoded text. If None, the value is determined by the clean_up_tokenization_spaces attribute of
                the XLNetTokenizer instance.
            spaces_between_special_tokens (bool): A flag indicating whether to include spaces between special tokens
                in the decoded text. Defaults to True.
            **kwargs: Additional keyword arguments. 'use_source_tokenizer' is a supported argument to control the
                use of the source tokenizer during decoding.

        Returns:
            str: The decoded string representation of the token IDs.

        Raises:
            None
        """
        self._decode_use_source_tokenizer = kwargs.pop("use_source_tokenizer", False)

        filtered_tokens = self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens)

        # To avoid mixing byte-level and unicode for byte-level BPT
        # we need to build string separately for added tokens and byte-level tokens
        # cf. https://github.com/huggingface/transformers/issues/1133
        sub_texts = []
        current_sub_text = []
        for token in filtered_tokens:
            if skip_special_tokens and token in self.all_special_ids:
                continue
            if token in self.added_tokens_encoder:
                if current_sub_text:
                    sub_texts.append(self.convert_tokens_to_string(current_sub_text))
                    current_sub_text = []
                sub_texts.append(token)
            else:
                current_sub_text.append(token)
        if current_sub_text:
            sub_texts.append(self.convert_tokens_to_string(current_sub_text))

        # Mimic the behavior of the Rust tokenizer:
        # By default, there are no spaces between special tokens
        text = "".join(sub_texts)

        clean_up_tokenization_spaces = (
            clean_up_tokenization_spaces
            if clean_up_tokenization_spaces is not None
            else self.clean_up_tokenization_spaces
        )
        if clean_up_tokenization_spaces:
            clean_text = self.clean_up_tokenization(text)
            return clean_text
        else:
            return text

    def build_inputs_with_special_tokens(
            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
    ) -> List[int]:
        """
        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
        adding special tokens. An XLNet sequence has the following format:

        - single sequence: `X <sep> <cls>`
        - pair of sequences: `A <sep> B <sep> <cls>`

        Args:
            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[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
        """
        sep = [self.sep_token_id]
        cls = [self.cls_token_id]
        if token_ids_1 is None:
            return token_ids_0 + sep + cls
        return token_ids_0 + sep + token_ids_1 + sep + cls

    def get_special_tokens_mask(
            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None,
            already_has_special_tokens: bool = False
    ) -> List[int]:
        """
        Retrieve 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` method.

        Args:
            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`):
                Whether or not the token list is already formatted with special tokens for the model.

        Returns:
            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
        """
        if already_has_special_tokens:
            return super().get_special_tokens_mask(
                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
            )

        if token_ids_1 is not None:
            return ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1, 1]
        return ([0] * len(token_ids_0)) + [1, 1]

    def create_token_type_ids_from_sequences(
            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
    ) -> List[int]:
        """
        Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet
        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`, this method only returns the first portion of the mask (0s).

        Args:
            token_ids_0 (`List[int]`):
                List of IDs.
            token_ids_1 (`List[int]`, *optional*):
                Optional second list of IDs for sequence pairs.

        Returns:
            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
        """
        sep = [self.sep_token_id]
        cls_segment_id = [2]

        if token_ids_1 is None:
            return len(token_ids_0 + sep) * [0] + cls_segment_id
        return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1] + cls_segment_id

    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
        """
        Save the vocabulary of the XLNetTokenizer.

        Args:
            self (XLNetTokenizer): An instance of the XLNetTokenizer class.
            save_directory (str): The directory path where the vocabulary will be saved.
            filename_prefix (Optional[str]): An optional prefix for the filename of the saved vocabulary.
                Defaults to None.

        Returns:
            Tuple[str]: A tuple containing the path to the saved vocabulary file.

        Raises:
            FileNotFoundError: If the specified save_directory does not exist.
            PermissionError: If the specified save_directory is not accessible for writing.

        Note:
            - The saved vocabulary file will be named as per the following format:
            '<filename_prefix>-vocab.txt' if filename_prefix is provided, otherwise 'vocab.txt'.
            - If the provided save_directory is the same as the current vocabulary file's directory and
            the vocabulary file already exists, it will be copied to the save_directory.
            - If the current vocabulary file does not exist, a new vocabulary file will be created in the
            save_directory using the serialized model from the sp_model attribute of the tokenizer.

        Example:
            ```python
            >>> tokenizer = XLNetTokenizer()
            >>> save_dir = '/path/to/save'
            >>> prefix = 'english'
            >>> vocab_file = tokenizer.save_vocabulary(save_dir, prefix)
            >>> print(f"Vocabulary saved at: {vocab_file}")
            ```
        """
        if not os.path.isdir(save_directory):
            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
            return
        out_vocab_file = os.path.join(
            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
        )

        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
            copyfile(self.vocab_file, out_vocab_file)
        elif not os.path.isfile(self.vocab_file):
            with open(out_vocab_file, "wb") as fi:
                content_spiece_model = self.sp_model.serialized_model_proto()
                fi.write(content_spiece_model)

        return (out_vocab_file,)

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.vocab_size property

Returns the vocabulary size of the XLNetTokenizer.

PARAMETER DESCRIPTION
self

An instance of the XLNetTokenizer class.

TYPE: XLNetTokenizer

RETURNS DESCRIPTION
int

The vocabulary size of the tokenizer.

This method calculates and returns the size of the vocabulary used by the XLNetTokenizer. The vocabulary size is determined by the number of unique tokens present in the tokenizer's sp_model.

Note

The vocabulary size represents the number of distinct tokens that the tokenizer can recognize and encode.

Example 
>>> tokenizer = XLNetTokenizer()
>>> size = tokenizer.vocab_size()
>>> print(size)
32000

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.__getstate__()

Method 'getstate' in the class 'XLNetTokenizer'.

PARAMETER DESCRIPTION
self

The instance of the XLNetTokenizer class. Represents the current XLNetTokenizer object.

TYPE: XLNetTokenizer

RETURNS DESCRIPTION
dict

This method returns a dictionary representing the state of the XLNetTokenizer object. The 'sp_model' key in the dictionary is set to None before returning.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
def __getstate__(self):
    """
    Method '__getstate__' in the class 'XLNetTokenizer'.

    Args:
        self (XLNetTokenizer): The instance of the XLNetTokenizer class.
            Represents the current XLNetTokenizer object.

    Returns:
        dict: This method returns a dictionary representing the state of the XLNetTokenizer object.
            The 'sp_model' key in the dictionary is set to None before returning.

    Raises:
        None.
    """
    state = self.__dict__.copy()
    state["sp_model"] = None
    return state

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.__init__(vocab_file, do_lower_case=False, remove_space=True, keep_accents=False, bos_token='<s>', eos_token='</s>', unk_token='<unk>', sep_token='<sep>', pad_token='<pad>', cls_token='<cls>', mask_token='<mask>', additional_special_tokens=['<eop>', '<eod>'], sp_model_kwargs=None, **kwargs)

Initialize an XLNetTokenizer object.

PARAMETER DESCRIPTION
vocab_file

Path to the vocabulary file.

TYPE: str

do_lower_case

Whether to lowercase the input tokens. Defaults to False.

TYPE: bool DEFAULT: False

remove_space

Whether to remove spaces in the input tokens. Defaults to True.

TYPE: bool DEFAULT: True

keep_accents

Whether to keep accents in the input tokens. Defaults to False.

TYPE: bool DEFAULT: False

bos_token

Beginning of sentence token. Defaults to ''.

TYPE: str DEFAULT: '<s>'

eos_token

End of sentence token. Defaults to ''.

TYPE: str DEFAULT: '</s>'

unk_token

Unknown token. Defaults to ''.

TYPE: str DEFAULT: '<unk>'

sep_token

Separator token. Defaults to ''.

TYPE: str DEFAULT: '<sep>'

pad_token

Padding token. Defaults to ''.

TYPE: str DEFAULT: '<pad>'

cls_token

Classification token. Defaults to ''.

TYPE: str DEFAULT: '<cls>'

mask_token

Mask token. Defaults to ''.

TYPE: str DEFAULT: '<mask>'

additional_special_tokens

Additional special tokens to include. Defaults to ['', ''].

TYPE: list DEFAULT: ['<eop>', '<eod>']

sp_model_kwargs

SentencePiece model keyword arguments. Defaults to None.

TYPE: Dict[str, Any] DEFAULT: None

**kwargs

Additional keyword arguments.

DEFAULT: {}

RETURNS DESCRIPTION
None

None
RAISES DESCRIPTION
TypeError

If the mask_token is not a string.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
def __init__(
        self,
        vocab_file,
        do_lower_case=False,
        remove_space=True,
        keep_accents=False,
        bos_token="<s>",
        eos_token="</s>",
        unk_token="<unk>",
        sep_token="<sep>",
        pad_token="<pad>",
        cls_token="<cls>",
        mask_token="<mask>",
        additional_special_tokens=["<eop>", "<eod>"],
        sp_model_kwargs: Optional[Dict[str, Any]] = None,
        **kwargs,
) -> None:
    """
    Initialize an XLNetTokenizer object.

    Args:
        vocab_file (str): Path to the vocabulary file.
        do_lower_case (bool, optional): Whether to lowercase the input tokens. Defaults to False.
        remove_space (bool, optional): Whether to remove spaces in the input tokens. Defaults to True.
        keep_accents (bool, optional): Whether to keep accents in the input tokens. Defaults to False.
        bos_token (str, optional): Beginning of sentence token. Defaults to '<s>'.
        eos_token (str, optional): End of sentence token. Defaults to '</s>'.
        unk_token (str, optional): Unknown token. Defaults to '<unk>'.
        sep_token (str, optional): Separator token. Defaults to '<sep>'.
        pad_token (str, optional): Padding token. Defaults to '<pad>'.
        cls_token (str, optional): Classification token. Defaults to '<cls>'.
        mask_token (str, optional): Mask token. Defaults to '<mask>'.
        additional_special_tokens (list, optional): Additional special tokens to include.
            Defaults to ['<eop>', '<eod>'].
        sp_model_kwargs (Dict[str, Any], optional): SentencePiece model keyword arguments. Defaults to None.
        **kwargs: Additional keyword arguments.

    Returns:
        None

    Raises:
        TypeError: If the mask_token is not a string.
    """
    # Mask token behave like a normal word, i.e. include the space before it
    mask_token = AddedToken(mask_token, lstrip=True, special=True) if isinstance(mask_token, str) else mask_token

    self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs

    self.do_lower_case = do_lower_case
    self.remove_space = remove_space
    self.keep_accents = keep_accents
    self.vocab_file = vocab_file

    self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
    self.sp_model.Load(vocab_file)

    super().__init__(
        do_lower_case=do_lower_case,
        remove_space=remove_space,
        keep_accents=keep_accents,
        bos_token=bos_token,
        eos_token=eos_token,
        unk_token=unk_token,
        sep_token=sep_token,
        pad_token=pad_token,
        cls_token=cls_token,
        mask_token=mask_token,
        additional_special_tokens=additional_special_tokens,
        sp_model_kwargs=self.sp_model_kwargs,
        **kwargs,
    )

    self._pad_token_type_id = 3

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.__setstate__(d)

This method setstate is defined in the class XLNetTokenizer and is used to set the internal state of the object based on the provided dictionary 'd'.

PARAMETER DESCRIPTION
self

The instance of the XLNetTokenizer class.

TYPE: XLNetTokenizer

d

A dictionary containing the state information to be set. The keys and values in the dictionary are used to update the internal state of the XLNetTokenizer object.

TYPE: dict

RETURNS DESCRIPTION

None. This method does not return any value.
RAISES DESCRIPTION
None

However, potential exceptions that could be raised include:

  • AttributeError: If the 'sp_model_kwargs' attribute is not found within the XLNetTokenizer object.
  • TypeError: If the provided 'd' parameter is not a dictionary.
  • Other exceptions related to the SentencePieceProcessor object creation or loading process may be raised from the spm.SentencePieceProcessor constructor or Load method.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
def __setstate__(self, d):
    """
    This method __setstate__ is defined in the class XLNetTokenizer and is used to set the internal state of the
    object based on the provided dictionary 'd'.

    Args:
        self (XLNetTokenizer): The instance of the XLNetTokenizer class.
        d (dict): A dictionary containing the state information to be set. The keys and values in the dictionary
            are used to update the internal state of the XLNetTokenizer object.

    Returns:
        None. This method does not return any value.

    Raises:
        None:
            However, potential exceptions that could be raised include:

            - AttributeError: If the 'sp_model_kwargs' attribute is not found within the XLNetTokenizer object.
            - TypeError: If the provided 'd' parameter is not a dictionary.
            - Other exceptions related to the SentencePieceProcessor object creation or loading process may be
            raised from the spm.SentencePieceProcessor constructor or Load method.
    """
    self.__dict__ = d

    # for backward compatibility
    if not hasattr(self, "sp_model_kwargs"):
        self.sp_model_kwargs = {}

    self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
    self.sp_model.Load(self.vocab_file)

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.build_inputs_with_special_tokens(token_ids_0, token_ids_1=None)

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

  • single sequence: X <sep> <cls>
  • pair of sequences: A <sep> B <sep> <cls>
PARAMETER DESCRIPTION
token_ids_0

List of IDs to which the special tokens will be added.

TYPE: `List[int]`

token_ids_1

Optional second list of IDs for sequence pairs.

TYPE: `List[int]`, *optional* DEFAULT: None

RETURNS DESCRIPTION
List[int]

List[int]: List of input IDs with the appropriate special tokens.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
def build_inputs_with_special_tokens(
        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
    """
    Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
    adding special tokens. An XLNet sequence has the following format:

    - single sequence: `X <sep> <cls>`
    - pair of sequences: `A <sep> B <sep> <cls>`

    Args:
        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[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
    """
    sep = [self.sep_token_id]
    cls = [self.cls_token_id]
    if token_ids_1 is None:
        return token_ids_0 + sep + cls
    return token_ids_0 + sep + token_ids_1 + sep + cls

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.convert_tokens_to_string(tokens)

Converts a sequence of tokens (strings for sub-words) in a single string.

Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
368
369
370
371
def convert_tokens_to_string(self, tokens):
    """Converts a sequence of tokens (strings for sub-words) in a single string."""
    out_string = "".join(tokens).replace(SPIECE_UNDERLINE, " ").strip()
    return out_string

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.create_token_type_ids_from_sequences(token_ids_0, token_ids_1=None)

Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet 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, this method only returns the first portion of the mask (0s).

PARAMETER DESCRIPTION
token_ids_0

List of IDs.

TYPE: `List[int]`

token_ids_1

Optional second list of IDs for sequence pairs.

TYPE: `List[int]`, *optional* DEFAULT: None

RETURNS DESCRIPTION
List[int]

List[int]: List of token type IDs according to the given sequence(s).
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
def create_token_type_ids_from_sequences(
        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
    """
    Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet
    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`, this method only returns the first portion of the mask (0s).

    Args:
        token_ids_0 (`List[int]`):
            List of IDs.
        token_ids_1 (`List[int]`, *optional*):
            Optional second list of IDs for sequence pairs.

    Returns:
        `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
    """
    sep = [self.sep_token_id]
    cls_segment_id = [2]

    if token_ids_1 is None:
        return len(token_ids_0 + sep) * [0] + cls_segment_id
    return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1] + cls_segment_id

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.get_special_tokens_mask(token_ids_0, token_ids_1=None, already_has_special_tokens=False)

Retrieve 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 method.

PARAMETER DESCRIPTION
token_ids_0

List of IDs.

TYPE: `List[int]`

token_ids_1

Optional second list of IDs for sequence pairs.

TYPE: `List[int]`, *optional* DEFAULT: None

already_has_special_tokens

Whether or not the token list is already formatted with special tokens for the model.

TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False

RETURNS DESCRIPTION
List[int]

List[int]: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
def get_special_tokens_mask(
        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None,
        already_has_special_tokens: bool = False
) -> List[int]:
    """
    Retrieve 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` method.

    Args:
        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`):
            Whether or not the token list is already formatted with special tokens for the model.

    Returns:
        `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
    """
    if already_has_special_tokens:
        return super().get_special_tokens_mask(
            token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
        )

    if token_ids_1 is not None:
        return ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1, 1]
    return ([0] * len(token_ids_0)) + [1, 1]

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.get_vocab()

Returns the vocabulary of the XLNetTokenizer.

PARAMETER DESCRIPTION
self

The instance of the XLNetTokenizer class.
RETURNS DESCRIPTION
dict

A dictionary containing the vocabulary of the XLNetTokenizer. The keys of the dictionary are the tokens, and the values are their corresponding indices.
Example 
>>> tokenizer = XLNetTokenizer()
>>> tokenizer.get_vocab()
{'<s>': 0, '<pad>': 1, '</s>': 2, '<unk>': 3, '<mask>': 4, 'hello': 5, 'world': 6}
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
def get_vocab(self):
    """
    Returns the vocabulary of the XLNetTokenizer.

    Args:
        self: The instance of the XLNetTokenizer class.

    Returns:
        dict:
            A dictionary containing the vocabulary of the XLNetTokenizer.
            The keys of the dictionary are the tokens, and the values are their corresponding indices.

    Raises:
        None.

    Example:
        ```python
        >>> tokenizer = XLNetTokenizer()
        >>> tokenizer.get_vocab()
        {'<s>': 0, '<pad>': 1, '</s>': 2, '<unk>': 3, '<mask>': 4, 'hello': 5, 'world': 6}
        ```
    """
    vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
    vocab.update(self.added_tokens_encoder)
    return vocab

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.preprocess_text(inputs)

This method preprocesses the input text according to the specified configuration settings.

PARAMETER DESCRIPTION
self

The instance of the XLNetTokenizer class.

TYPE: XLNetTokenizer

inputs

The input text to be preprocessed. It should be a string representation.

TYPE: str

RETURNS DESCRIPTION
str

The preprocessed text based on the applied configuration settings.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
def preprocess_text(self, inputs):
    """
    This method preprocesses the input text according to the specified configuration settings.

    Args:
        self (XLNetTokenizer): The instance of the XLNetTokenizer class.
        inputs (str): The input text to be preprocessed. It should be a string representation.

    Returns:
        str: The preprocessed text based on the applied configuration settings.

    Raises:
        None
    """
    if self.remove_space:
        outputs = " ".join(inputs.strip().split())
    else:
        outputs = inputs
    outputs = outputs.replace("``", '"').replace("''", '"')

    if not self.keep_accents:
        outputs = unicodedata.normalize("NFKD", outputs)
        outputs = "".join([c for c in outputs if not unicodedata.combining(c)])
    if self.do_lower_case:
        outputs = outputs.lower()

    return outputs

mindnlp.transformers.models.xlnet.tokenization_xlnet.XLNetTokenizer.save_vocabulary(save_directory, filename_prefix=None)

Save the vocabulary of the XLNetTokenizer.

PARAMETER DESCRIPTION
self

An instance of the XLNetTokenizer class.

TYPE: XLNetTokenizer

save_directory

The directory path where the vocabulary will be saved.

TYPE: str

filename_prefix

An optional prefix for the filename of the saved vocabulary. Defaults to None.

TYPE: Optional[str] DEFAULT: None

RETURNS DESCRIPTION
Tuple[str]

Tuple[str]: A tuple containing the path to the saved vocabulary file.
RAISES DESCRIPTION
FileNotFoundError

If the specified save_directory does not exist.
PermissionError

If the specified save_directory is not accessible for writing.
Note
  • The saved vocabulary file will be named as per the following format: '-vocab.txt' if filename_prefix is provided, otherwise 'vocab.txt'.
  • If the provided save_directory is the same as the current vocabulary file's directory and the vocabulary file already exists, it will be copied to the save_directory.
  • If the current vocabulary file does not exist, a new vocabulary file will be created in the save_directory using the serialized model from the sp_model attribute of the tokenizer.
Example 
>>> tokenizer = XLNetTokenizer()
>>> save_dir = '/path/to/save'
>>> prefix = 'english'
>>> vocab_file = tokenizer.save_vocabulary(save_dir, prefix)
>>> print(f"Vocabulary saved at: {vocab_file}")
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet.py 
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
    """
    Save the vocabulary of the XLNetTokenizer.

    Args:
        self (XLNetTokenizer): An instance of the XLNetTokenizer class.
        save_directory (str): The directory path where the vocabulary will be saved.
        filename_prefix (Optional[str]): An optional prefix for the filename of the saved vocabulary.
            Defaults to None.

    Returns:
        Tuple[str]: A tuple containing the path to the saved vocabulary file.

    Raises:
        FileNotFoundError: If the specified save_directory does not exist.
        PermissionError: If the specified save_directory is not accessible for writing.

    Note:
        - The saved vocabulary file will be named as per the following format:
        '<filename_prefix>-vocab.txt' if filename_prefix is provided, otherwise 'vocab.txt'.
        - If the provided save_directory is the same as the current vocabulary file's directory and
        the vocabulary file already exists, it will be copied to the save_directory.
        - If the current vocabulary file does not exist, a new vocabulary file will be created in the
        save_directory using the serialized model from the sp_model attribute of the tokenizer.

    Example:
        ```python
        >>> tokenizer = XLNetTokenizer()
        >>> save_dir = '/path/to/save'
        >>> prefix = 'english'
        >>> vocab_file = tokenizer.save_vocabulary(save_dir, prefix)
        >>> print(f"Vocabulary saved at: {vocab_file}")
        ```
    """
    if not os.path.isdir(save_directory):
        logger.error(f"Vocabulary path ({save_directory}) should be a directory")
        return
    out_vocab_file = os.path.join(
        save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
    )

    if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
        copyfile(self.vocab_file, out_vocab_file)
    elif not os.path.isfile(self.vocab_file):
        with open(out_vocab_file, "wb") as fi:
            content_spiece_model = self.sp_model.serialized_model_proto()
            fi.write(content_spiece_model)

    return (out_vocab_file,)

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast

Tokenization classes for XLNet model.

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast

Bases: PreTrainedTokenizerFast

Construct a "fast" XLNet tokenizer (backed by HuggingFace's tokenizers library). Based on Unigram.

This tokenizer inherits from [PreTrainedTokenizerFast] which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

PARAMETER DESCRIPTION
vocab_file

SentencePiece file (generally has a .spm extension) that contains the vocabulary necessary to instantiate a tokenizer.

TYPE: `str` DEFAULT: None

do_lower_case

Whether to lowercase the input when tokenizing.

TYPE: `bool`, *optional*, defaults to `True` DEFAULT: False

remove_space

Whether to strip the text when tokenizing (removing excess spaces before and after the string).

TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True

keep_accents

Whether to keep accents when tokenizing.

TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False

bos_token

The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.

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.

TYPE: `str`, *optional*, defaults to `"<s>"` DEFAULT: '<s>'

eos_token

The end of sequence token.

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

TYPE: `str`, *optional*, defaults to `"</s>"` DEFAULT: '</s>'

unk_token

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.

TYPE: `str`, *optional*, defaults to `"<unk>"` DEFAULT: '<unk>'

sep_token

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.

TYPE: `str`, *optional*, defaults to `"<sep>"` DEFAULT: '<sep>'

pad_token

The token used for padding, for example when batching sequences of different lengths.

TYPE: `str`, *optional*, defaults to `"<pad>"` DEFAULT: '<pad>'

cls_token

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.

TYPE: `str`, *optional*, defaults to `"<cls>"` DEFAULT: '<cls>'

mask_token

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.

TYPE: `str`, *optional*, defaults to `"<mask>"` DEFAULT: '<mask>'

additional_special_tokens

Additional special tokens used by the tokenizer.

TYPE: `List[str]`, *optional*, defaults to `["<eop>", "<eod>"]` DEFAULT: ['<eop>', '<eod>']

ATTRIBUTE DESCRIPTION
sp_model

The SentencePiece processor that is used for every conversion (string, tokens and IDs).

TYPE: `SentencePieceProcessor`

Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet_fast.py 
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
class XLNetTokenizerFast(PreTrainedTokenizerFast):
    """
    Construct a "fast" XLNet tokenizer (backed by HuggingFace's *tokenizers* library). Based on
    [Unigram](https://huggingface.co/docs/tokenizers/python/latest/components.html?highlight=unigram#models).

    This tokenizer inherits from [`PreTrainedTokenizerFast`] which contains most of the main methods. Users should
    refer to this superclass for more information regarding those methods.

    Args:
        vocab_file (`str`):
            [SentencePiece](https://github.com/google/sentencepiece) file (generally has a .spm extension) that
            contains the vocabulary necessary to instantiate a tokenizer.
        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.
        bos_token (`str`, *optional*, defaults to `"<s>"`):
            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.

            <Tip>

            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`.

            </Tip>

        eos_token (`str`, *optional*, defaults to `"</s>"`):
            The end of sequence token.

            <Tip>

            When building a sequence using special tokens, this is not the token that is used for the end of sequence.
            The token used is the `sep_token`.

            </Tip>

        unk_token (`str`, *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.
        sep_token (`str`, *optional*, defaults to `"<sep>"`):
            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 (`str`, *optional*, defaults to `"<pad>"`):
            The token used for padding, for example when batching sequences of different lengths.
        cls_token (`str`, *optional*, defaults to `"<cls>"`):
            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 (`str`, *optional*, defaults to `"<mask>"`):
            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 `["<eop>", "<eod>"]`):
            Additional special tokens used by the tokenizer.

    Attributes:
        sp_model (`SentencePieceProcessor`):
            The *SentencePiece* processor that is used for every conversion (string, tokens and IDs).
    """
    vocab_files_names = VOCAB_FILES_NAMES
    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
    padding_side = "left"
    slow_tokenizer_class = XLNetTokenizer

    def __init__(
        self,
        vocab_file=None,
        tokenizer_file=None,
        do_lower_case=False,
        remove_space=True,
        keep_accents=False,
        bos_token="<s>",
        eos_token="</s>",
        unk_token="<unk>",
        sep_token="<sep>",
        pad_token="<pad>",
        cls_token="<cls>",
        mask_token="<mask>",
        additional_special_tokens=["<eop>", "<eod>"],
        **kwargs,
    ):
        """
        __init__

        Initializes an instance of the XLNetTokenizerFast class.

        Args:
            self: The instance of the class.
            vocab_file (str, optional): The path to the vocabulary file. Defaults to None.
            tokenizer_file (str, optional): The path to the tokenizer file. Defaults to None.
            do_lower_case (bool, optional): Whether to convert tokens to lowercase. Defaults to False.
            remove_space (bool, optional): Whether to remove spaces from tokens. Defaults to True.
            keep_accents (bool, optional): Whether to keep accents in tokens. Defaults to False.
            bos_token (str, optional): The beginning of sentence token. Defaults to '<s>'.
            eos_token (str, optional): The end of sentence token. Defaults to '</s>'.
            unk_token (str, optional): The unknown token. Defaults to '<unk>'.
            sep_token (str, optional): The separator token. Defaults to '<sep>'.
            pad_token (str, optional): The padding token. Defaults to '<pad>'.
            cls_token (str, optional): The classification token. Defaults to '<cls>'.
            mask_token (str or AddedToken, optional): The mask token. Defaults to '<mask>'.
            additional_special_tokens (list, optional): Additional special tokens. Defaults to ['<eop>', '<eod>'].
            **kwargs: Additional keyword arguments.

        Returns:
            None.

        Raises:
            ValueError: If invalid input is provided for the parameters.
            TypeError: If the input type for the parameters is incorrect.
        """
        # Mask token behave like a normal word, i.e. include the space before it
        mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token

        super().__init__(
            vocab_file=vocab_file,
            tokenizer_file=tokenizer_file,
            do_lower_case=do_lower_case,
            remove_space=remove_space,
            keep_accents=keep_accents,
            bos_token=bos_token,
            eos_token=eos_token,
            unk_token=unk_token,
            sep_token=sep_token,
            pad_token=pad_token,
            cls_token=cls_token,
            mask_token=mask_token,
            additional_special_tokens=additional_special_tokens,
            **kwargs,
        )

        self._pad_token_type_id = 3
        self.do_lower_case = do_lower_case
        self.remove_space = remove_space
        self.keep_accents = keep_accents
        self.vocab_file = vocab_file

    @property
    def can_save_slow_tokenizer(self) -> bool:
        """
        Checks if the slow tokenizer can be saved.

        Args:
            self (XLNetTokenizerFast): An instance of the XLNetTokenizerFast class.

        Returns:
            bool: Returns True if the slow tokenizer can be saved, False otherwise.

        Raises:
            None.

        The 'can_save_slow_tokenizer' method checks if the slow tokenizer can be saved by verifying the existence of
        the vocabulary file. It returns a boolean value indicating whether the slow tokenizer can be saved or not.
        If the 'vocab_file' attribute is not set or if the file does not exist, the method returns False. Otherwise,
        it returns True.

        Note that this method does not raise any exceptions.
        """
        return os.path.isfile(self.vocab_file) if self.vocab_file else False

    def build_inputs_with_special_tokens(
        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
    ) -> List[int]:
        """
        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
        adding special tokens. An XLNet sequence has the following format:

        - single sequence: `X <sep> <cls>`
        - pair of sequences: `A <sep> B <sep> <cls>`

        Args:
            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[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
        """
        sep = [self.sep_token_id]
        cls = [self.cls_token_id]
        if token_ids_1 is None:
            return token_ids_0 + sep + cls
        return token_ids_0 + sep + token_ids_1 + sep + cls

    def create_token_type_ids_from_sequences(
        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
    ) -> List[int]:
        """
        Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet
        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`, this method only returns the first portion of the mask (0s).

        Args:
            token_ids_0 (`List[int]`):
                List of IDs.
            token_ids_1 (`List[int]`, *optional*):
                Optional second list of IDs for sequence pairs.

        Returns:
            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
        """
        sep = [self.sep_token_id]
        cls_segment_id = [2]

        if token_ids_1 is None:
            return len(token_ids_0 + sep) * [0] + cls_segment_id
        return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1] + cls_segment_id

    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
        """Save the vocabulary for a fast tokenizer to a specified directory.

        Args:
            self (XLNetTokenizerFast): The instance of the XLNetTokenizerFast class.
            save_directory (str): The directory path where the vocabulary will be saved.
            filename_prefix (Optional[str]): A prefix for the filename. Defaults to None. 

        Returns:
            Tuple[str]: A tuple containing the path to the saved vocabulary file.

        Raises:
            ValueError: If the fast tokenizer does not have the necessary information to save the vocabulary for
                a slow tokenizer.
            OSError: If the save_directory does not exist or is not a valid directory.
            IOError: If an error occurs while copying the vocabulary file to the specified directory.
        """
        if not self.can_save_slow_tokenizer:
            raise ValueError(
                "Your fast tokenizer does not have the necessary information to save the vocabulary for a slow "
                "tokenizer."
            )

        if not os.path.isdir(save_directory):
            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
            return
        out_vocab_file = os.path.join(
            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
        )

        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
            copyfile(self.vocab_file, out_vocab_file)

        return (out_vocab_file,)

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.can_save_slow_tokenizer: bool property

Checks if the slow tokenizer can be saved.

PARAMETER DESCRIPTION
self

An instance of the XLNetTokenizerFast class.

TYPE: XLNetTokenizerFast

RETURNS DESCRIPTION
bool

Returns True if the slow tokenizer can be saved, False otherwise.

TYPE: bool

The 'can_save_slow_tokenizer' method checks if the slow tokenizer can be saved by verifying the existence of the vocabulary file. It returns a boolean value indicating whether the slow tokenizer can be saved or not. If the 'vocab_file' attribute is not set or if the file does not exist, the method returns False. Otherwise, it returns True.

Note that this method does not raise any exceptions.

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.__init__(vocab_file=None, tokenizer_file=None, do_lower_case=False, remove_space=True, keep_accents=False, bos_token='<s>', eos_token='</s>', unk_token='<unk>', sep_token='<sep>', pad_token='<pad>', cls_token='<cls>', mask_token='<mask>', additional_special_tokens=['<eop>', '<eod>'], **kwargs)

init

Initializes an instance of the XLNetTokenizerFast class.

PARAMETER DESCRIPTION
self

The instance of the class.
vocab_file

The path to the vocabulary file. Defaults to None.

TYPE: str DEFAULT: None

tokenizer_file

The path to the tokenizer file. Defaults to None.

TYPE: str DEFAULT: None

do_lower_case

Whether to convert tokens to lowercase. Defaults to False.

TYPE: bool DEFAULT: False

remove_space

Whether to remove spaces from tokens. Defaults to True.

TYPE: bool DEFAULT: True

keep_accents

Whether to keep accents in tokens. Defaults to False.

TYPE: bool DEFAULT: False

bos_token

The beginning of sentence token. Defaults to ''.

TYPE: str DEFAULT: '<s>'

eos_token

The end of sentence token. Defaults to ''.

TYPE: str DEFAULT: '</s>'

unk_token

The unknown token. Defaults to ''.

TYPE: str DEFAULT: '<unk>'

sep_token

The separator token. Defaults to ''.

TYPE: str DEFAULT: '<sep>'

pad_token

The padding token. Defaults to ''.

TYPE: str DEFAULT: '<pad>'

cls_token

The classification token. Defaults to ''.

TYPE: str DEFAULT: '<cls>'

mask_token

The mask token. Defaults to ''.

TYPE: str or AddedToken DEFAULT: '<mask>'

additional_special_tokens

Additional special tokens. Defaults to ['', ''].

TYPE: list DEFAULT: ['<eop>', '<eod>']

**kwargs

Additional keyword arguments.

DEFAULT: {}

RETURNS DESCRIPTION

None.
RAISES DESCRIPTION
ValueError

If invalid input is provided for the parameters.
TypeError

If the input type for the parameters is incorrect.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet_fast.py 
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
def __init__(
    self,
    vocab_file=None,
    tokenizer_file=None,
    do_lower_case=False,
    remove_space=True,
    keep_accents=False,
    bos_token="<s>",
    eos_token="</s>",
    unk_token="<unk>",
    sep_token="<sep>",
    pad_token="<pad>",
    cls_token="<cls>",
    mask_token="<mask>",
    additional_special_tokens=["<eop>", "<eod>"],
    **kwargs,
):
    """
    __init__

    Initializes an instance of the XLNetTokenizerFast class.

    Args:
        self: The instance of the class.
        vocab_file (str, optional): The path to the vocabulary file. Defaults to None.
        tokenizer_file (str, optional): The path to the tokenizer file. Defaults to None.
        do_lower_case (bool, optional): Whether to convert tokens to lowercase. Defaults to False.
        remove_space (bool, optional): Whether to remove spaces from tokens. Defaults to True.
        keep_accents (bool, optional): Whether to keep accents in tokens. Defaults to False.
        bos_token (str, optional): The beginning of sentence token. Defaults to '<s>'.
        eos_token (str, optional): The end of sentence token. Defaults to '</s>'.
        unk_token (str, optional): The unknown token. Defaults to '<unk>'.
        sep_token (str, optional): The separator token. Defaults to '<sep>'.
        pad_token (str, optional): The padding token. Defaults to '<pad>'.
        cls_token (str, optional): The classification token. Defaults to '<cls>'.
        mask_token (str or AddedToken, optional): The mask token. Defaults to '<mask>'.
        additional_special_tokens (list, optional): Additional special tokens. Defaults to ['<eop>', '<eod>'].
        **kwargs: Additional keyword arguments.

    Returns:
        None.

    Raises:
        ValueError: If invalid input is provided for the parameters.
        TypeError: If the input type for the parameters is incorrect.
    """
    # Mask token behave like a normal word, i.e. include the space before it
    mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token

    super().__init__(
        vocab_file=vocab_file,
        tokenizer_file=tokenizer_file,
        do_lower_case=do_lower_case,
        remove_space=remove_space,
        keep_accents=keep_accents,
        bos_token=bos_token,
        eos_token=eos_token,
        unk_token=unk_token,
        sep_token=sep_token,
        pad_token=pad_token,
        cls_token=cls_token,
        mask_token=mask_token,
        additional_special_tokens=additional_special_tokens,
        **kwargs,
    )

    self._pad_token_type_id = 3
    self.do_lower_case = do_lower_case
    self.remove_space = remove_space
    self.keep_accents = keep_accents
    self.vocab_file = vocab_file

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.build_inputs_with_special_tokens(token_ids_0, token_ids_1=None)

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

  • single sequence: X <sep> <cls>
  • pair of sequences: A <sep> B <sep> <cls>
PARAMETER DESCRIPTION
token_ids_0

List of IDs to which the special tokens will be added.

TYPE: `List[int]`

token_ids_1

Optional second list of IDs for sequence pairs.

TYPE: `List[int]`, *optional* DEFAULT: None

RETURNS DESCRIPTION
List[int]

List[int]: List of input IDs with the appropriate special tokens.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet_fast.py 
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
def build_inputs_with_special_tokens(
    self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
    """
    Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
    adding special tokens. An XLNet sequence has the following format:

    - single sequence: `X <sep> <cls>`
    - pair of sequences: `A <sep> B <sep> <cls>`

    Args:
        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[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
    """
    sep = [self.sep_token_id]
    cls = [self.cls_token_id]
    if token_ids_1 is None:
        return token_ids_0 + sep + cls
    return token_ids_0 + sep + token_ids_1 + sep + cls

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.create_token_type_ids_from_sequences(token_ids_0, token_ids_1=None)

Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet 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, this method only returns the first portion of the mask (0s).

PARAMETER DESCRIPTION
token_ids_0

List of IDs.

TYPE: `List[int]`

token_ids_1

Optional second list of IDs for sequence pairs.

TYPE: `List[int]`, *optional* DEFAULT: None

RETURNS DESCRIPTION
List[int]

List[int]: List of token type IDs according to the given sequence(s).
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet_fast.py 
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
def create_token_type_ids_from_sequences(
    self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
    """
    Create a mask from the two sequences passed to be used in a sequence-pair classification task. An XLNet
    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`, this method only returns the first portion of the mask (0s).

    Args:
        token_ids_0 (`List[int]`):
            List of IDs.
        token_ids_1 (`List[int]`, *optional*):
            Optional second list of IDs for sequence pairs.

    Returns:
        `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
    """
    sep = [self.sep_token_id]
    cls_segment_id = [2]

    if token_ids_1 is None:
        return len(token_ids_0 + sep) * [0] + cls_segment_id
    return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1] + cls_segment_id

mindnlp.transformers.models.xlnet.tokenization_xlnet_fast.XLNetTokenizerFast.save_vocabulary(save_directory, filename_prefix=None)

Save the vocabulary for a fast tokenizer to a specified directory.

PARAMETER DESCRIPTION
self

The instance of the XLNetTokenizerFast class.

TYPE: XLNetTokenizerFast

save_directory

The directory path where the vocabulary will be saved.

TYPE: str

filename_prefix

A prefix for the filename. Defaults to None.

TYPE: Optional[str] DEFAULT: None

RETURNS DESCRIPTION
Tuple[str]

Tuple[str]: A tuple containing the path to the saved vocabulary file.
RAISES DESCRIPTION
ValueError

If the fast tokenizer does not have the necessary information to save the vocabulary for a slow tokenizer.
OSError

If the save_directory does not exist or is not a valid directory.
IOError

If an error occurs while copying the vocabulary file to the specified directory.
Source code in mindnlp/transformers/models/xlnet/tokenization_xlnet_fast.py 
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
    """Save the vocabulary for a fast tokenizer to a specified directory.

    Args:
        self (XLNetTokenizerFast): The instance of the XLNetTokenizerFast class.
        save_directory (str): The directory path where the vocabulary will be saved.
        filename_prefix (Optional[str]): A prefix for the filename. Defaults to None. 

    Returns:
        Tuple[str]: A tuple containing the path to the saved vocabulary file.

    Raises:
        ValueError: If the fast tokenizer does not have the necessary information to save the vocabulary for
            a slow tokenizer.
        OSError: If the save_directory does not exist or is not a valid directory.
        IOError: If an error occurs while copying the vocabulary file to the specified directory.
    """
    if not self.can_save_slow_tokenizer:
        raise ValueError(
            "Your fast tokenizer does not have the necessary information to save the vocabulary for a slow "
            "tokenizer."
        )

    if not os.path.isdir(save_directory):
        logger.error(f"Vocabulary path ({save_directory}) should be a directory")
        return
    out_vocab_file = os.path.join(
        save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
    )

    if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
        copyfile(self.vocab_file, out_vocab_file)

    return (out_vocab_file,)