is_last should not be checked forValueFunction

[Josh00-Lu]

Thanks for your wonderful project.

I think, here, is_last should not be checked for ValueFunction, since there's no nn.Identity() here for the last block:

        for ind, (dim_in, dim_out) in enumerate(in_out):
            is_last = ind >= (num_resolutions - 1)

            self.blocks.append(nn.ModuleList([
                ResidualTemporalBlock(dim_in, dim_out, kernel_size=5, embed_dim=time_dim, horizon=horizon),
                ResidualTemporalBlock(dim_out, dim_out, kernel_size=5, embed_dim=time_dim, horizon=horizon),
                Downsample1d(dim_out)
            ]))

            if not is_last: # Should delete this line
                horizon = horizon // 2

Otherwise, the dimension may not be correctly handled:

e.g. For horizon=32 and dim_mults=(1, 4, 8), a dimension-mismatch error would appear:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (1x96 and 160x64)

        for ind, (dim_in, dim_out) in enumerate(in_out):
            is_last = ind >= (num_resolutions - 1)

            self.blocks.append(nn.ModuleList([
                ResidualTemporalBlock(dim_in, dim_out, kernel_size=5, embed_dim=time_dim, horizon=horizon),
                ResidualTemporalBlock(dim_out, dim_out, kernel_size=5, embed_dim=time_dim, horizon=horizon),
                Downsample1d(dim_out)
            ]))

            horizon = horizon // 2

This bug will not appear when horizon is small (0 // 2 =0), but will appear for large horizon.