RuntimeError: expected scalar type Float but found Half

def __init__(self, chi, cho):
    super(DeformConv, self).__init__()
    self.actf = nn.Sequential(
        nn.BatchNorm2d(cho, momentum=BN_MOMENTUM),
        nn.ReLU(inplace=True)
    )
    self.conv = DCN(chi, cho, kernel_size=(3, 3), stride=1, padding=1, dilation=1, deformable_groups=1)
def forward(self, x):
    x = self.conv(x)
    x = self.actf(x)
    return x

My model exsits a DCN module which compiled by c++. when I use amp.initialize(model, optimizer, opt_level="O1"), RuntimeError has happened(expected scalar type Float but found Half) in x=self.conv(x). I try to use x=self.conv(x.float()) to convert type, but not useful.

Start at the shown DeformConv.forward call and inspect how the C++-compiled DCN module handles the tensor dtype under amp.initialize with opt_level="O1". Compare the input and extension expectations around x=self.conv(x), including the attempted float conversion. Done means the reported RuntimeError is resolved for this mixed-precision path; no specific files or tests are named.