Exclude pre-trained word embedding from parameters

1

My model looks as below:

class ERModeler(nn.Module):
    def __init__(self, vocab_size, embedding_dim, weInput):
        super(ERModeler, self).__init__()
        self.embeddings = nn.Embedding.from_pretrained(weInput)
        self.embeddings.weight.requires_grad = False
        self.linear = nn.Linear(1, 2)

When I check the the parameters, it looks surprising to me that the first layer of parameter is the weights of embedding. I expect it’s there only when self.embeddings.weight.requires_grad = True

model = ERModeler(VOC_SIZE, EMBEDDING_DIM, weInput)
print(list(model.parameters())[0])