JSON strings cannot generate with "\" character in them

[posionus]

Describe the issue as clearly as possible:

I have fine tuned a model to generate json. I am using outlines with vLLM to force the model to generate correct json schema. Outlines is failing to generate:

{
  "cost": null,
  "description": "Net Equity Attributable to Holders of Redeemable Units (\"Net Equity\")",
  "footnotes": [],
  "percent_net_assets": 100.0,
  "value": 3767045854
}

Specifically, it is failing to generate the "(\"Net Equity\")" in the description field.

The issue seems to be with this line

outlines/outlines/fsm/json_schema.py

Line 13 in 7723ce8

STRING_INNER = r'([^"\\\x00-\x1f\x7f-\x9f]|\\\\)'

It is prohibiting any \ character inside of the string.

"Net Equity Attributable to Holders of Redeemable Units (\"Net Equity\")" is a valid json string. I have tested parsing the json above using python json.loads. It should be able to generate using outlines.

Steps/code to reproduce the bug:

My schema:

class StatementInvestmentsTotal(BaseModel):
    cost: Optional[float]
    description: Optional[str]
    footnotes: list[str]
    percent_net_assets: Optional[float]
    value: Optional[float]

Expected result:

Outlines should allow generating backslashes in json strings

Error message:

No response

Outlines/Python version information:

Version information

``` (command output here) ```

Context for the issue:

No response

[lapp0]

Edit: sorry replied to the wrong thread, I'll take a look at this soon.

[lapp0]

@posionus any interest verifying and creating a PR for this fix? I have a bit on my backlog.

>>> interegular.parse_pattern(json_schema.STRING).to_fsm().accepts(r'"my escaped string: \" okay"')
False

>>> STRING_INNER = r'(?:[^"\\]|\\.)*?'
>>> STRING = r'"' + STRING_INNER + r'"'
>>> interegular.parse_pattern(STRING).to_fsm().accepts(r'"my escaped string:  \" okay"')
True