Restrict backpropagation in specific indices in nn.Embedding

nlp
1

In nn.Embedding can we provide an option so that specific indices are never updated? Like a mask for example?