conv_bias // 2 should be correct.
self.conv1d_x = nn.Conv1d(
in_channels=self.d_inner // 2,
out_channels=self.d_inner // 2,
bias=conv_bias,
kernel_size=d_conv,
groups=self.d_inner // 2,
**factory_kwargs,
)
self.conv1d_z = nn.Conv1d(
in_channels=self.d_inner // 2,
out_channels=self.d_inner // 2,
bias=conv_bias,
kernel_size=d_conv,
groups=self.d_inner // 2,
**factory_kwargs,
)
conv_bias // 2 should be correct.