| import torch | |
| from torch import nn | |
| from transformers import AutoModel | |
| class PhoBERTPairABSA(nn.Module): | |
| """Pair-ABSA model: Predicts sentiment for a specific topic in a sentence""" | |
| def __init__(self, base_model="vinai/phobert-base", num_cls=4, dropout=0.2): | |
| super().__init__() | |
| self.backbone = AutoModel.from_pretrained(base_model) | |
| hidden_size = self.backbone.config.hidden_size | |
| self.classifier = nn.Sequential( | |
| nn.Dropout(dropout), | |
| nn.Linear(hidden_size, hidden_size), | |
| nn.GELU(), | |
| nn.LayerNorm(hidden_size), | |
| nn.Dropout(dropout), | |
| nn.Linear(hidden_size, num_cls) | |
| ) | |
| def forward(self, input_ids, attention_mask): | |
| out = self.backbone(input_ids=input_ids, attention_mask=attention_mask) | |
| cls = out.last_hidden_state[:, 0, :] | |
| logits = self.classifier(cls) | |
| return logits | |