You are trying to self.log() but it is not managed by the Trainer control flow

[nian-liu]

Hello everyone,

I am new to pytorch-lighting and I have the following error with a simple classification code:

I am using self.log() like this:

This error still occurs even I comment all these self.log() lines. Any suggestion will be appreciated. Thanks!

[rohitgr7]

hey @nian-liu !

are you sure these are all the self.log calls you have in your code? since it looks correct.

if possible can you share your code? or a reproducible script using https://colab.research.google.com/github/PyTorchLightning/pytorch-lightning/blob/master/pl_examples/bug_report/bug_report_model.ipynb

[nian-liu]

Hi @rohitgr7,

I found where the problem is.
Using the flag log_gpu_memory='all' in pl.Trainer leads to conducting self._log_gpus_metrics() in logger_connector.py, which may be "not managed by the Trainer control flow". After removing this flag, the error is gone.

[rohitgr7]

this flag was removed recently. You can use DeviceStatsMonitor instead to log GPU memory stats.