important code error: losses["loss"].detach() in train_util.py line 246

[Frank21st]

please check the losses variable

        if isinstance(self.schedule_sampler, LossAwareSampler):
            self.schedule_sampler.update_with_local_losses(
                t, losses["loss"].detach()
            )
        losses = losses1[0]
        sample = losses1[1]

[heha102]

I met the same situation. Do you know how to solve it now?

[Frank21st]

waiting for the author to reply...

[WuJunde]

you can simply put line 248 in front of it. actually this block will not be run in the current version

[Frank21st]

losses["loss"].detach() ---> losses1[0].detach() ?