Bug description

When using the AdvancedProfiler to create and persist a profile of a training, it is not possible to have a ModelCheckpoint which monitors a variable with a slash in it, e.g. "val/JaccardIndex". The training run will result in a "No such file or directory" error.

What version are you seeing the problem on?

v2.5

Reproduced in studio

No response

How to reproduce the bug

profiler_dir.mkdir(parents=True, exist_ok=True)
profiler = AdvancedProfiler(dirpath=profiler_dir, filename="perf_logs", dump_stats=True)

checkpoint_callback = ModelCheckpoint(
            filename="epoch={epoch}-step={step}-val_iou={val/JaccardIndex:.2f}",
            auto_insert_metric_name=False,
            verbose=True,
            monitor="val/JaccardIndex",
            mode="max",
            save_last="link",
            save_top_k=training_config.save_top_k,
        )
L.Trainer(
        callbacks=[checkpoint_callback],
        profiler=profiler,
        ...
    )

Error messages and logs

It seems like the action_name of the ModelCheckpoint profiler is action_name = "[Callback]ModelCheckpoint{'monitor': 'val/JaccardIndex', 'mode': 'max', 'every_n"+73

FileNotFoundError: [Errno 2] No such file or directory: ".../profiler/fit-perf_logs-[Callback]ModelCheckpoint{'monitor': 'val/JaccardIndex', 'mode': 'max', 'every_n_train_steps': 0,'every_n_epochs': 1, 'train_time_interval': None}.setup.prof"

Environment

More info

It seems like having the configuration of the callback in the filename is desired behavior: #19703

However, maybe this should be done differently. Maybe by sanitizing the action_name or by doing something similar like ModelCheckpoint with auto_insert_metric_name.

cc @ethanwharris @lantiga