code
stringlengths
3
6.57k
__init__(self, config)
super()
__init__()
nn.ModuleList([BertLayer(config)
range(config.num_hidden_layers)
nn.ModuleList([BertHighway(config)
range(config.num_hidden_layers)
range(config.num_hidden_layers)
set_early_exit_entropy(self, x)
if (type(x)
or (type(x)
range(len(self.early_exit_entropy)
init_highway_pooler(self, pooler)
pooler.state_dict()
highway.pooler.state_dict()
items()
param.copy_(loaded_model[name])
enumerate(self.layer)
entropy(highway_logits)
hidden_states(?)
HighwayException(new_output, i + 1)
exiting (DeeBERT)
DeeBertModel(BertPreTrainedModel)
__init__(self, config)
super()
__init__(config)
BertEmbeddings(config)
DeeBertEncoder(config)
BertPooler(config)
self.init_weights()
init_highway_pooler(self)
self.encoder.init_highway_pooler(self.pooler)
get_input_embeddings(self)
set_input_embeddings(self, value)
_prune_heads(self, heads_to_prune)
heads_to_prune.items()
attention.prune_heads(heads)
add_start_docstrings_to_model_forward(BERT_INPUTS_DOCSTRING)
tuple(torch.FloatTensor)
configuration (:class:`~transformers.BertConfig`)
last_hidden_state (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)
pooler_output (:obj:`torch.FloatTensor`: of shape :obj:`(batch_size, hidden_size)
sequence (classification token)
prediction (classification)
hidden_states (:obj:`tuple(torch.FloatTensor)
attentions (:obj:`tuple(torch.FloatTensor)
highway_exits (:obj:`tuple(tuple(torch.Tensor)
results (total length: number of layers)
ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
input_ids.size()
inputs_embeds.size()
ValueError("You have to specify either input_ids or inputs_embeds")
torch.ones(input_shape, device=device)
torch.ones(input_shape, device=device)
torch.zeros(input_shape, dtype=torch.long, device=device)
self.get_extended_attention_mask(attention_mask, input_shape, device)
encoder_attention_mask.dim()
encoder_attention_mask.dim()
next(self.parameters()
self.get_head_mask(head_mask, self.config.num_hidden_layers)
self.pooler(sequence_output)
HighwayException(Exception)
__init__(self, message, exit_layer)
BertHighway(nn.Module)
from (the output of one non-final BertLayer in BertEncoder)
to (cross-entropy computation in BertForSequenceClassification)
__init__(self, config)
super()
__init__()
BertPooler(config)
nn.Dropout(config.hidden_dropout_prob)
nn.Linear(config.hidden_size, config.num_labels)
forward(self, encoder_outputs)
self.pooler(pooler_input)
self.dropout(pooled_output)
self.classifier(pooled_output)
Model (with early exiting - DeeBERT)
DeeBertForSequenceClassification(BertPreTrainedModel)
__init__(self, config)
super()
__init__(config)
DeeBertModel(config)
nn.Dropout(config.hidden_dropout_prob)
nn.Linear(config.hidden_size, self.config.num_labels)
self.init_weights()
add_start_docstrings_to_model_forward(BERT_INPUTS_DOCSTRING)
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)
computed (Mean-Square loss)
computed (Cross-Entropy)
tuple(torch.FloatTensor)
configuration (:class:`~transformers.BertConfig`)
loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)
Classification (or regression if config.num_labels==1)
logits (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, config.num_labels)
Classification (or regression if config.num_labels==1)
scores (before SoftMax)
hidden_states (:obj:`tuple(torch.FloatTensor)
attentions (:obj:`tuple(torch.FloatTensor)
highway_exits (:obj:`tuple(tuple(torch.Tensor)
results (total length: number of layers)