File size: 770 Bytes
ceee6db |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 |
import torch
import tqdm
from torch import nn
from transformers import MT5EncoderModel, MT5PreTrainedModel
class MT5EncoderWithProjection(MT5PreTrainedModel):
def __init__(self, config):
super().__init__(config)
self.config = config
self.mt5_encoder = MT5EncoderModel(config)
self.projection = nn.Linear(config.d_model, config.d_model, bias=False)
self.post_init()
def forward(self, **input_args):
hidden_states = self.mt5_encoder(**input_args).last_hidden_state
mask = input_args['attention_mask']
batch_embeddings = torch.sum(hidden_states * mask[:, :, None], dim=1) / torch.sum(mask, dim=1)[:, None]
batch_embeddings = self.projection(batch_embeddings)
return batch_embeddings
|