Solution:
In model/yolo.py file

        for mi, s in zip(m.m, m.stride):  # from
            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
            b[:, 4] += math.log(8/(640/s) ** 2)  # obj (8 objects per 640 image)
            b[:, 5:] += math.log(0.6/(m.nc - 0.99)) if cf is None else torch.log(cf/cf.sum())  # cls
            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)

Add with torch.no_grad(): as follows

        for mi, s in zip(m.m, m.stride):  # from
            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
            with torch.no_grad():
                b[:, 4] += math.log(8/(640/s) ** 2)  # obj (8 objects per 640 image)
                b[:, 5:] += math.log(0.6/(m.nc - 0.99)) if cf is None else torch.log(cf/cf.sum())  # cls
            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)

——> The root cause is to add:

with torch.no_grad()：