modeling_t5_regression.py

from transformers import T5EncoderModel, T5Config, PreTrainedModel
import torch.nn as nn
import torch

class T5RegressionModel(PreTrainedModel):
    config_class = T5Config

    def __init__(self, config, d_model=None):
        super().__init__(config)
        self.encoder = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_uniref50")
        hidden_dim = d_model if d_model is not None else config.d_model
        self.regression_head = nn.Linear(hidden_dim, 1)

    def forward(self, input_ids=None, attention_mask=None, labels=None, **kwargs):
        encoder_outputs = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
        hidden_states = encoder_outputs.last_hidden_state
        pooled_output = hidden_states[:, -1, :]
        logits = self.regression_head(pooled_output).squeeze(-1)

        loss = None
        if labels is not None:
            labels = labels.float()
            loss = nn.MSELoss()(logits, labels)

        return {"loss": loss, "logits": logits}