When resuming with BackboneFinetuning callback, backbone parameters has requires_grad=False

[sour4bh]

When I resume from a checkpoint that has the backbone unfrozen (i.e. current_epoch > unfreeze_backbone_at_epoch), the parameters of the backbone are once again frozen.
I checked this in the following manner:

def training_epoch_end(self, _):
    for name, param in self.named_parameters():
        print(param.requires_grad, name)

Although I was able to verify that the optimizer upon resuming from checkpoint does contain two param groups which is expected.

[rohitgr7]

can you share a minimal example to reproduce. will help debug this