File size: 2,596 Bytes
d759971 f3ed620 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 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 |
from transformers import PretrainedConfig
from abc import ABCMeta
import torch
from transformers.pytorch_utils import nn
from transformers import BertModel, BertConfig
import torch
import torch.nn.functional as F
from transformers import PreTrainedModel
from transformers.modeling_outputs import SequenceClassifierOutput
from transformers import PretrainedConfig
class BertLSTMConfig(PretrainedConfig):
model_type = "bertLSTMForSequenceClassification"
def __init__(self,
num_classes=2,
embed_dim=768,
num_layers=12,
hidden_dim_lstm=256, # New parameter for LSTM
dropout_rate=0.1,
**kwargs):
super().__init__(**kwargs)
self.num_classes = num_classes
self.embed_dim = embed_dim
self.num_layers = num_layers
self.hidden_dim_lstm = hidden_dim_lstm # Assign LSTM hidden dimension
self.dropout_rate = dropout_rate
self.id2label = {
0: "fake",
1: "true",
}
self.label2id = {
"fake": 0,
"true": 1,
}
class BertLSTMForSequenceClassification(PreTrainedModel, metaclass=ABCMeta):
config_class = BertLSTMConfig
def __init__(self, config):
super(BertLSTMForSequenceClassification, self).__init__(config)
self.num_classes = config.num_classes
self.embed_dim = config.embed_dim
self.num_layers = config.num_layers
self.hidden_dim_lstm = config.hidden_dim_lstm
self.dropout = nn.Dropout(config.dropout_rate)
self.bert = BertModel.from_pretrained('bert-base-uncased')
print("BERT Model Loaded")
self.lstm = nn.LSTM(self.embed_dim, self.hidden_dim_lstm, batch_first=True, num_layers=3)
self.fc = nn.Linear(self.hidden_dim_lstm, self.num_classes)
def forward(self, input_ids, attention_mask, token_type_ids, labels=None):
bert_output = self.bert(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)
pooled_output = bert_output.pooler_output # Use the pooled output for classification
out, _ = self.lstm(pooled_output.unsqueeze(1))
out = self.dropout(out[:, -1, :])
logits = self.fc(out)
loss = None
if labels is not None:
loss = F.cross_entropy(logits, labels)
out = SequenceClassifierOutput(
loss=loss,
logits=logits,
hidden_states=bert_output.hidden_states,
attentions=bert_output.attentions,
)
return out |