Self.log with multiple Optimizers errors out in 1.1.0 but works in 1.0.8

[blisc]

🐛 Bug

self.log results in the following error:

  File "/opt/conda/lib/python3.6/site-packages/pytorch_lightning/trainer/connectors/logger_connector/epoch_result_store.py", line 207, in auto_reduce_results_on_epoch_end
    opt_outputs = epoch_metrics[opt_idx]

when the following conditions are satisfied:

• There are two optimizers
• 1 of the optimizers does not log anything in 1 training epoch

Note, that this bug is not present in v1.0.8 and was introduced in 1.1.0

Please reproduce using the BoringModel and post here

https://colab.research.google.com/drive/1MkMmuzTmZU2hkPjoEQ8Mwd4qSh168QKn?usp=sharing

To Reproduce

See notebook

Expected behavior

The same behavious as 1.0.8.

Environment

See notebook.

Additional context

See notebook.

[edenlightning]

@tchaton mind taking a look?