log_gpu_memory='min_max' leads to error in parsing metrics keys

[cosw0t]

🐛 Bug

log_gpu_memory='min_max' raises:

/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py in _log_gpus_metrics(self)
    227     def _log_gpus_metrics(self):
    228         for key, mem in self.gpus_metrics.items():
--> 229             gpu_id = int(key.split("/")[0].split(":")[1])
    230             if gpu_id in self.trainer.accelerator_connector.parallel_device_ids:
    231                 self.trainer.lightning_module.log(key, mem, prog_bar=False, logger=True, on_step=True, on_epoch=False)

To Reproduce

https://colab.research.google.com/drive/13Q_YzybNXDRL7gBZsluIRqfBIq21G3Et?usp=sharing