|
import torch.nn as nn |
|
from transformers import BertModel |
|
import pytorch_lightning as pl |
|
|
|
BERT_MODEL_NAME = 'alger-ia/dziribert' |
|
class Dialect_Detection(pl.LightningModule): |
|
def __init__(self, n_classes): |
|
super().__init__() |
|
self.bert = BertModel.from_pretrained(BERT_MODEL_NAME) |
|
self.classifier = nn.Linear(self.bert.config.hidden_size, n_classes) |
|
self.criterion = nn.CrossEntropyLoss() |
|
|
|
def forward(self, input_ids, attention_mask, labels=None): |
|
output = self.bert(input_ids, attention_mask) |
|
output = self.classifier(output.pooler_output) |
|
|
|
if labels is not None: |
|
loss = self.criterion(output, labels) |
|
return loss, output |
|
|
|
return output |