ValueGatedAttention.Config(
                channels_in=16,
                channels_out=16,
                num_heads=2,
                channels_head=8,
                gate_channels=4,
                norm_qk=RMSNorm.Config(
                        channels_in=8,
                        channels_out=8,
                        eps=1e-06,
                        elementwise_affine=False,
                        device=None,
                        dtype=None
                ),
                init_weight=<function priml.model.init.unit_fan_in_uniform at 0xdefacedeface>,
                kernel=SdpaCausal.Config(),
                window=4,
                window_pattern='SSSL',
                max_seq_len=-1,
                gated=True,
                depth_index=()
        )
