why the value of µ_t0 and α_t0 be always assigned to zero vectors?

[AlphaMind123]

ml-tarflow/transformer_flow.py

Line 172 in 313120e

x = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1)

Why should µ_t0 and α_t0 be assigned to zero vectors after passing through MetaBlock？because of the following code：

x = self.proj_out(x) # (b, t, 32)
x = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) #
if self.nvp: # True
xa, xb = x.chunk(2, dim=-1) # alpha, miu