Escape backslash in ignore regex pattern

[jubueche]

Here the string describing the regex pattern contains backslashes.
Python raises a SyntaxWarning: invalid escape sequence '\.' as a result.
Should these backslashes be escaped?

I don't think it's a bug but clarification would be nice.

[brian-dellabetta]

Hi @jubueche , I don't see that warning raised anywhere when running the example (Python 3.10, latest llm-compressor main). Can you provide your logs to show when it is displaying?

Both seem to work, re.compile seems to understand that the \ in regex needs to be escaped in python during compilation:

>>> r=re.compile("model\.layer")
re.compile('model\\.layer')
>>> re.match("model\.layer", "model.layer")
<re.Match object; span=(0, 11), match='model.layer'>
>>> re.match("model\\.layer", "model.layer")
<re.Match object; span=(0, 11), match='model.layer'>

[jubueche]

Hi. You're not seeing this because this got bumped to SyntaxWarning from DeprecationWarning (which is per-se slienced) in Python 3.12. I'd suggest to adapt the examples to properly escape the backslashes then.
See here: https://docs.python.org/3/whatsnew/3.12.html#other-language-changes

[brian-dellabetta]

I worry about adding escape characters because regexes are already difficult for all users to grasp, but instead of

"re:model\.vision_tower.*"

we could do

r"re:model\.vision_tower.*"

so that the escape character is added when python parses a raw string into a string?

[jubueche]

Yes I guess that works 👍