SoftCap.Config(
                channels_in=4,
                channels_out=4,
                cap=2.0,
                inner=Linear.Config(
                        channels_in=4,
                        channels_out=4,
                        bias=False,
                        depth_index=(),
                        device=None,
                        dtype=None,
                        init_weight=<function priml.model.init.kaiming_uniform at 0xdefacedeface>,
                        init_bias=<function torch.nn.init.zeros_ at 0xdefacedeface>,
                        shard=None
                ),
                dtype=torch.float32
        )
