should I use mask in pytorch encoder only model?

Question

I want to use transformer to build one translation model, and I have achieve one encoder-decoder model by pytorch.

but now I wounder could I use encoder only to achieve this, and I have write some code

class Transformer(nn.Module):
    def __init__(self):
        super(Transformer, self).__init__()
        self.device = device
        self.src_emb = nn.Embedding(src_vocab_size, d_model)
        self.transformer_encoder = nn.TransformerEncoder(
            nn.TransformerEncoderLayer(d_model=d_model, nhead=n_heads, dim_feedforward=d_ff, batch_first=True),
            num_layers=n_layers,
        )
        self.projection = nn.Linear(d_model, tgt_vocab_size, bias=True)

    def forward(self, src):
        src_padding_mask = (src == 0)
        # src_mask = nn.Transformer.generate_square_subsequent_mask(src.size(1)).bool().to(self.device)
        memory = self.src_emb(src)
        # memory = self.transformer_encoder(memory, src_key_padding_mask=src_padding_mask, mask=src_mask)
        memory = self.transformer_encoder(memory, src_key_padding_mask=src_padding_mask)
        output = self.projection(memory)
        return output.view(-1, output.shape[-1])

but i don't know if I should use src_mask or not,

by reading other author answer, I know that we can use the nn.encoder with mask to achieve gpt like model.

so i think the nn.encoder only with mask may be correct, but i don't know without that mask this code is right or not.

that's all, thanks.

could u tell me something about this?