surkovvv's picture
Create model.py
6d2f9da
raw
history blame
787 Bytes
from transformers import DistilBertModel
import torch
class DistillBERTClass(torch.nn.Module):
def __init__(self):
super(DistillBERTClass, self).__init__()
self.l1 = DistilBertModel.from_pretrained("distilbert-base-uncased")
self.pre_classifier = torch.nn.Linear(768, 512)
self.dropout = torch.nn.Dropout(0.3)
self.classifier = torch.nn.Linear(512, 126)
def forward(self, input_ids, attention_mask):
output_1 = self.l1(input_ids=input_ids, attention_mask=attention_mask)
hidden_state = output_1[0]
pooler = hidden_state[:, 0]
pooler = self.pre_classifier(pooler)
pooler = torch.nn.ReLU()(pooler)
pooler = self.dropout(pooler)
output = self.classifier(pooler)
return output