How do you deal with matrix/tensor local various in RichHandler?

[ml-w]

Rich is super for error messages, but if I am dealing with pytorch where there's tons of tensors in a local scope, the reporting become very long (see image). I've tried to set max_length...etc options but I don't they are designed for the purpose. I wonder how everyone is dealing with this problem? How can I limit the number of rows each locals can take?

[TomJGooding]

There's a similar discussion (and related PR) you might find interesting here:

• Workaround & Discussion: Customizing object representations in tracebacks (e.g., for NumPy arrays) #3774