activation

eMultiheadAttention(in_rep, num_heads[, ...])

Drop-in replacement for torch.nn.MultiheadAttention that preserves G-equivariance.