Expose more arguments for PyTorchProfiler

[tensorcopy]

Description & Motivation

There are only limited arguments for PyTorch profiler exposed to the user. Some other arguments might be useful, such as max_name_column_width when the path is long and truncated.

Pitch

Expose more arguments of PyTorch profiler such as max_name_column_width.

Alternatives

Customize a lightning profiler to get a new PyTorch profiler.

Additional context

No response

cc @Borda @carmocca @nbcsm @guotuofeng

[carmocca]

The class definition accepts **profiler_kwargs. So you should be able to pass this already: https://github.com/Lightning-AI/lightning/blob/a5c43d3b2b80f5fc769d7ed0ea511c0bd6733c6b/src/lightning/pytorch/profilers/pytorch.py#L243

[tensorcopy]

The class definition accepts **profiler_kwargs. So you should be able to pass this already:

Thanks for taking a look! But I don't think passing arguments like max_name_column_width will work. It should be added here similar to _sort_by_key and _row_limit.

https://github.com/Lightning-AI/lightning/blob/a5c43d3b2b80f5fc769d7ed0ea511c0bd6733c6b/src/lightning/pytorch/profilers/pytorch.py#L488

[carmocca]

I see! Then you could add table_kwargs to the signature