epoch/step validation with MLflowLogger

[kota-iizuka]

I'm trying to perform a 100 step experiment on a dataset with 1 epoch = 22 steps and logging every 10 steps by MLFlowLogger, but it doesn't work.

The settings for the Trainer are as follows

   max_epochs: null
   max_steps: 100
   val_check_interval: 10
   check_val_every_n_epoch: null

I am using mlflow logger and the output image is as below.

Judging from the graph, it appears that even though check_val_every_n_epoch is set to None, validation is occurring at the end of each epoch and at the end of trainer.fit(), and the epoch number is incorrectly recorded as step.

Is there a setting to resolve this? Or do I need to modify pytorch-lightning itself?