Avoid adding None loss values in training_epoch_end (#7772)

Author: carmocca

CHANGELOG.md

@@ -161,6 +161,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed `ProgressBar` pickling after calling `trainer.predict` ([#7608](https://github.com/PyTorchLightning/pytorch-lightning/pull/7608))
 
 
+- Fixed `None` loss keys getting added in `training_epoch_end` when using manual optimization and not returning a loss ([#7772](https://github.com/PyTorchLightning/pytorch-lightning/pull/7772))
+
+
 - Fixed print errors in `ProgressBar` when `trainer.fit` is not called ([#7674](https://github.com/PyTorchLightning/pytorch-lightning/pull/7674))
 
 

pytorch_lightning/trainer/training_loop.py

@@ -390,7 +390,8 @@ def _prepare_outputs(
 
                 for tbptt_output in batch_outputs:
                     out = tbptt_output.extra
-                    out['loss'] = tbptt_output.minimize
+                    if tbptt_output.minimize is not None:
+                        out['loss'] = tbptt_output.minimize.detach()
                     processed_tbptt_outputs.append(out)
 
                 # if there was only one tbptt step then we can collapse that dimension