modeling#

class ElectraModel(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

The bare Electra Model transformer outputting raw hidden-states.

This model inherits from PretrainedModel. Refer to the superclass documentation for the generic methods.

This model is also a Paddle paddle.nn.Layer subclass. Use it as a regular Paddle Layer and refer to the Paddle documentation for all matter related to general usage and behavior.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

get_input_embeddings()[source]#

get input embedding of model

Returns:

embedding of model

Return type:

nn.Embedding

set_input_embeddings(value)[source]#

set new input embedding for model

Parameters:

value (Embedding) – the new embedding of model

Raises:

NotImplementedError – Model has not implement set_input_embeddings method

forward(input_ids, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, past_key_values=None, use_cache=None, output_attentions=False, output_hidden_states=False, return_dict=False)[source]#

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

Parameters:

  • input_ids (Tensor) – Indices of input sequence tokens in the vocabulary. They are numerical representations of tokens that build the input sequence. Its data type should be int64 and it has a shape of [batch_size, sequence_length].

  • token_type_ids (Tensor, optional) –

    Segment token indices to indicate different portions of the inputs. Selected in the range [0, type_vocab_size - 1]. If type_vocab_size is 2, which means the inputs have two portions. Indices can either be 0 or 1:

    • 0 corresponds to a sentence A token,

    • 1 corresponds to a sentence B token.

    Its data type should be int64 and it has a shape of [batch_size, sequence_length]. Defaults to None, which means we don’t add segment embeddings.

  • position_ids (Tensor, optional) – Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, max_position_embeddings - 1]. Shape as (batch_size, num_tokens) and dtype as int64. Defaults to None.

  • attention_mask (Tensor, optional) – Mask used in multi-head attention to avoid performing attention on to some unwanted positions, usually the paddings or the subsequent positions. Its data type can be int, float and bool. When the data type is bool, the masked tokens have False values and the others have True values. When the data type is int, the masked tokens have 0 values and the others have 1 values. When the data type is float, the masked tokens have -INF values and the others have 0 values. It is a tensor with shape broadcasted to [batch_size, num_attention_heads, sequence_length, sequence_length]. Defaults to None, which means nothing needed to be prevented attention to.

  • inputs_embeds (Tensor, optional) – Instead of passing input_ids you can choose to directly pass an embedded representation. This is useful for use cases such as P-Tuning, where you want more control over how to convert input_ids indices into the embedding space. Its data type should be float32 and it has a shape of [batch_size, sequence_length, embedding_size].

  • past_key_values (tuple(tuple(Tensor)), optional) – Precomputed key and value hidden states of the attention blocks of each layer. This can be used to speedup auto-regressive decoding for generation tasks or to support use cases such as Prefix-Tuning where vectors are prepended to each attention layer. The length of tuple equals to the number of layers, and each tuple having 2 tensors of shape (batch_size, num_heads, past_key_values_length, embed_size_per_head)) If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • use_cache (bool, optional) – If set to True, past_key_values key value states are returned. Defaults to None.

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tensor encoder_outputs, which is the output at the last layer of the model. Its data type should be float32 and has a shape of [batch_size, sequence_length, hidden_size].

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraModel, ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraModel.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
output = model(**inputs)
class ElectraPretrainedModel(*args, **kwargs)[source]#

Bases: PretrainedModel

An abstract class for pretrained Electra models. It provides Electra related model_config_file, pretrained_init_configuration, resource_files_names, pretrained_resource_files_map, base_model_prefix for downloading and loading pretrained models. See PretrainedModel for more details.

config_class#

alias of ElectraConfig

base_model_class#

alias of ElectraModel

class ElectraForTotalPretraining(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

Electra Model for pretraining tasks.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ElectraForTotalPretraining

get_input_embeddings()[source]#

get input embedding of model

Returns:

embedding of model

Return type:

nn.Embedding

get_output_embeddings()[source]#

To be overwrited for models with output embeddings

Returns:

the otuput embedding of model

Return type:

Optional[Embedding]

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, raw_input_ids=None, generator_labels=None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (list, optional) – See ElectraModel.

  • raw_input_ids (Tensor, optional) – Raw inputs used to get discriminator labels. Its data type should be int64 and it has a shape of [batch_size, sequence_length].

  • generator_labels (Tensor, optional) – Labels to compute the discriminator inputs. Its data type should be int64 and its shape is [batch_size, sequence_length]. The value for unmasked tokens should be -100 and value for masked tokens should be 0.

Returns:

Returns tuple (generator_logits, disc_logits, disc_labels, attention_mask).

With the fields:

  • generator_logits (Tensor):

    The scores of Electra Generator. Its data type should be int64 and its shape is [batch_size, sequence_length, vocab_size].

  • disc_logits (Tensor):

    The prediction result of replaced tokens. Its data type should be float32 and if batch_size>1, its shape is [batch_size, sequence_length], if batch_size=1, its shape is [sequence_length].

  • disc_labels (Tensor):

    The labels of electra discriminator. Its data type should be int32, and its shape is [batch_size, sequence_length].

  • attention_mask (Tensor):

    See ElectraModel. Its data type should be bool.

Return type:

tuple

class ElectraDiscriminator(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

The Electra Discriminator can detect the tokens that are replaced by the Electra Generator.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

forward(input_ids, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None)[source]#
Parameters:
Returns:

Returns tensor logits, the prediction result of replaced tokens. Its data type should be float32 and if batch_size>1, its shape is [batch_size, sequence_length], if batch_size=1, its shape is [sequence_length].

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraDiscriminator, ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraDiscriminator.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
logits = model(**inputs)
class ElectraGenerator(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

The Electra Generator will replace some tokens of the given sequence, it is trained as a masked language model.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

get_input_embeddings()[source]#

get input embedding of model

Returns:

embedding of model

Return type:

nn.Embedding

set_input_embeddings(value)[source]#

set new input embedding for model

Parameters:

value (Embedding) – the new embedding of model

Raises:

NotImplementedError – Model has not implement set_input_embeddings method

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, labels=None, output_attentions=False, output_hidden_states=False, return_dict=False)[source]#
Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (Tensor, optional) – See ElectraModel.

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tensor prediction_scores, the scores of Electra Generator. Its data type should be int64 and its shape is [batch_size, sequence_length, vocab_size].

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraGenerator, ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraGenerator.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
prediction_scores = model(**inputs)
class ElectraClassificationHead(config: ElectraConfig)[source]#

Bases: Layer

Perform sentence-level classification tasks.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

forward(features, **kwargs)[source]#

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

Parameters:

features (Tensor) – Input sequence, usually the sequence_output of electra model. Its data type should be float32 and its shape is [batch_size, sequence_length, hidden_size].

Returns:

Returns a tensor of the input text classification logits. Shape as [batch_size, num_labels] and dtype as float32.

Return type:

Tensor

class ElectraForSequenceClassification(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

Electra Model with a linear layer on top of the output layer, designed for sequence classification/regression tasks like GLUE tasks.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ElectraForSequenceClassification

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, labels=None, output_attentions: bool | None = None, output_hidden_states: bool | None = None, return_dict: bool | None = None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (list, optional) – See ElectraModel.

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tensor logits, a tensor of the input text classification logits. Shape as [batch_size, num_labels] and dtype as float32.

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraForSequenceClassification
from paddlenlp.transformers import ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraForSequenceClassification.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
logits = model(**inputs)
class ElectraForTokenClassification(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

Electra Model with a linear layer on top of the hidden-states output layer, designed for token classification tasks like NER tasks.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ElectraForTokenClassification

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, labels: Tensor | None = None, output_attentions: bool | None = None, output_hidden_states: bool | None = None, return_dict: bool | None = None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (list, optional) – See ElectraModel.

  • labels (Tensor 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)

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tensor logits, a tensor of the input token classification logits. Shape as [batch_size, sequence_length, num_labels] and dtype as float32.

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraForTokenClassification
from paddlenlp.transformers import ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraForTokenClassification.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
logits = model(**inputs)
class ElectraPretrainingCriterion(config: ElectraConfig)[source]#

Bases: Layer

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

forward(generator_prediction_scores, discriminator_prediction_scores, generator_labels, discriminator_labels, attention_mask)[source]#
Parameters:

  • generator_prediction_scores (Tensor) – The scores of masked token prediction. Its data type should be float32. and its shape is [batch_size, sequence_length, vocab_size].

  • discriminator_prediction_scores (Tensor) – The scores of masked token prediction. Its data type should be float32. and its shape is [batch_size, sequence_length] or [sequence length] if batch_size=1.

  • generator_labels (Tensor) – The labels of the generator, its dimensionality is equal to generator_prediction_scores. Its data type should be int64 and its shape is [batch_size, sequence_size, 1].

  • discriminator_labels (Tensor) – The labels of the discriminator, its dimensionality is equal to discriminator_prediction_scores. The labels should be numbers between 0 and 1. Its data type should be float32 and its shape is [batch_size, sequence_size] or [sequence length] if batch_size=1.

  • attention_mask (Tensor) – See ElectraModel.

Returns:

The pretraining loss, equals to weighted generator loss plus the weighted discriminator loss. Its data type should be float32 and its shape is [1].

Return type:

Tensor

class ElectraForMultipleChoice(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

Electra Model with a linear layer on top of the hidden-states output layer, designed for multiple choice tasks like RocStories/SWAG tasks.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ElectraForMultipleChoice

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, labels: Tensor | None = None, output_attentions: bool | None = None, output_hidden_states: bool | None = None, return_dict: bool | None = None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel and shape as [batch_size, num_choice, sequence_length].

  • token_type_ids (Tensor, optional) – See ElectraModel and shape as [batch_size, num_choice, sequence_length].

  • position_ids (Tensor, optional) – See ElectraModel and shape as [batch_size, num_choice, sequence_length].

  • attention_mask (list, optional) – See ElectraModel and shape as [batch_size, num_choice, sequence_length].

  • labels (Tensor 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)

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tensor reshaped_logits, a tensor of the multiple choice classification logits. Shape as [batch_size, num_choice] and dtype as float32.

Return type:

Tensor

Example

import paddle
from paddlenlp.transformers import ElectraForMultipleChoice, ElectraTokenizer
from paddlenlp.data import Pad, Dict

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraForMultipleChoice.from_pretrained('electra-small', num_choices=2)

data = [
    {
        "question": "how do you turn on an ipad screen?",
        "answer1": "press the volume button.",
        "answer2": "press the lock button.",
        "label": 1,
    },
    {
        "question": "how do you indent something?",
        "answer1": "leave a space before starting the writing",
        "answer2": "press the spacebar",
        "label": 0,
    },
]

text = []
text_pair = []
for d in data:
    text.append(d["question"])
    text_pair.append(d["answer1"])
    text.append(d["question"])
    text_pair.append(d["answer2"])

inputs = tokenizer(text, text_pair)
batchify_fn = lambda samples, fn=Dict(
    {
        "input_ids": Pad(axis=0, pad_val=tokenizer.pad_token_id),  # input_ids
        "token_type_ids": Pad(
            axis=0, pad_val=tokenizer.pad_token_type_id
        ),  # token_type_ids
    }
): fn(samples)
inputs = batchify_fn(inputs)

reshaped_logits = model(
    input_ids=paddle.to_tensor(inputs[0], dtype="int64"),
    token_type_ids=paddle.to_tensor(inputs[1], dtype="int64"),
)
print(reshaped_logits.shape)
# [2, 2]
class ElectraForQuestionAnswering(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

Electra Model with a linear layer on top of the hidden-states output to compute span_start_logits and span_end_logits, designed for question-answering tasks like SQuAD.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig used to construct ElectraForQuestionAnswering.

forward(input_ids, token_type_ids=None, position_ids=None, attention_mask=None, inputs_embeds=None, start_positions: Tensor | None = None, end_positions: Tensor | None = None, output_attentions: bool | None = None, output_hidden_states: bool | None = None, return_dict: bool | None = None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (list, optional) – See ElectraModel.

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

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

  • output_hidden_states (bool, optional) – Whether to return the hidden states of all layers. Defaults to False.

  • output_attentions (bool, optional) – Whether to return the attentions tensors of all attention layers. Defaults to False.

  • return_dict (bool, optional) – Whether to return a QuestionAnsweringModelOutput object. If False, the output will be a tuple of tensors. Defaults to False.

Returns:

Returns tuple (start_logits, end_logits).

With the fields:

  • start_logits (Tensor):

    A tensor of the input token classification logits, indicates the start position of the labelled span. Its data type should be float32 and its shape is [batch_size, sequence_length].

  • end_logits (Tensor):

    A tensor of the input token classification logits, indicates the end position of the labelled span. Its data type should be float32 and its shape is [batch_size, sequence_length].

Return type:

tuple

Example

import paddle
from paddlenlp.transformers import ElectraForQuestionAnswering, ElectraTokenizer

tokenizer = ElectraTokenizer.from_pretrained('electra-small')
model = ElectraForQuestionAnswering.from_pretrained('electra-small')

inputs = tokenizer("Welcome to use PaddlePaddle and PaddleNLP!")
inputs = {k:paddle.to_tensor([v]) for (k, v) in inputs.items()}
outputs = model(**inputs)

start_logits = outputs[0]
end_logits  = outputs[1]
ElectraForMaskedLM#

alias of ElectraGenerator

ElectraForPretraining#

alias of ElectraForTotalPretraining

class ErnieHealthForTotalPretraining(config: ElectraConfig)[source]#

Bases: ElectraForTotalPretraining

ERNIE-Health Model for pretraining task.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ElectraForMultipleChoice

sample_negatives_from_softmax(logits, raw_inputs, use_softmax_sample=True)[source]#

Sample K=5 non-original negative samples for candidate set.

Returns:

Returns tensor neg_samples_ids, a tensor of the negative samples of original inputs. Shape as ` [batch_size, sequence_length, K, vocab_size]` and dtype as int64.

Return type:

Tensor

forward(input_ids=None, token_type_ids=None, position_ids=None, attention_mask=None, raw_input_ids=None, generator_labels=None, return_dict: bool | None = None)[source]#

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

Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (list, optional) – See ElectraModel.

  • raw_input_ids (Tensor, optional) – Raw inputs used to get discriminator labels. Its data type should be int64 and it has a shape of [batch_size, sequence_length].

  • generator_labels (Tensor, optional) – Labels to compute the discriminator inputs. Its data type should be int64 and its shape is [batch_size, sequence_length]. The value for unmasked tokens should be -100 and value for masked tokens should be 0.

Returns:

Returns tuple (generator_logits, disc_logits, disc_labels, attention_mask).

With the fields:

  • generator_logits (Tensor):

    The scores of Electra Generator. Its data type should be int64 and its shape is [batch_size, sequence_length, vocab_size].

  • disc_logits (Tensor):

    The prediction result of replaced tokens. Its data type should be float32 and if batch_size>1, its shape is [batch_size, sequence_length], if batch_size=1, its shape is [sequence_length].

  • disc_labels (Tensor):

    The labels of electra discriminator. Its data type should be int32, and its shape is [batch_size, sequence_length].

  • attention_mask (Tensor):

    See ElectraModel. Its data type should be bool.

Return type:

tuple

class ErnieHealthPretrainingCriterion(config: ElectraConfig)[source]#

Bases: Layer

Parameters:

config (ElectraConfig) – An instance of ElectraConfig

forward(generator_logits, generator_labels, logits_rtd, logits_mts, logits_csp, discriminator_labels, attention_mask)[source]#
Parameters:

  • generator_logits (Tensor) – The scores of masked token prediction. Its data type should be float32. and its shape is [batch_size, sequence_length, vocab_size].

  • generator_labels (Tensor) – The labels of the generator, its dimensionality is equal to generator_prediction_scores. Its data type should be int64 and its shape is [batch_size, sequence_size, 1].

  • logits_rtd (Tensor) – The scores of masked token prediction. Its data type should be float32. and its shape is [batch_size, sequence_length] or [sequence length] if batch_size=1.

  • discriminator_labels (Tensor) – The labels of the discriminator, its dimensionality is equal to discriminator_prediction_scores. The labels should be numbers between 0 and 1. Its data type should be float32 and its shape is [batch_size, sequence_size] or [sequence length] if batch_size=1.

  • attention_mask (Tensor) – See ElectraModel.

Returns:

The pretraining loss, equals to weighted generator loss plus the weighted discriminator loss. Its data type should be float32 and its shape is [1].

Return type:

Tensor

class ErnieHealthDiscriminator(config: ElectraConfig)[source]#

Bases: ElectraPretrainedModel

The Discriminators in ERNIE-Health (https://arxiv.org/abs/2110.07244), including

  • token-level Replaced Token Detection (RTD) task

  • token-level Multi-Token Selection (MTS) task

  • sequence-level Contrastive Sequence Prediction (CSP) task.

Parameters:

config (ElectraConfig) – An instance of ElectraConfig to construct ErnieHealthDiscriminator

forward(input_ids, candidate_ids, token_type_ids=None, position_ids=None, attention_mask=None)[source]#
Parameters:

  • input_ids (Tensor) – See ElectraModel.

  • candidate_ids (Tensor) – The candidate indices of input sequence tokens in the vocabulary for MTS task. Its data type should be int64 and it has a shape of [batch_size, sequence_length].

  • token_type_ids (Tensor, optional) – See ElectraModel.

  • position_ids (Tensor, optional) – See ElectraModel.

  • attention_mask (Tensor, optional) – See ElectraModel.

Returns:

Returns list of tensors, the prediction results of RTD, MTS and CSP. The logits’ data type should be float32 and if batch_size > 1,

  • the shape of logits_rtd is [batch_size, sequence_length],

  • the shape of logits_mts is [batch_size, sequence_length, num_candidate],

  • the shape of logits_csp is [batch_size, 128].

If batch_size=1, the shapes are [sequence_length], [sequence_length, num_cadidate], [128], separately.

Return type:

Tensor