max_token & json utils

[Ali-Meh619]

PR Type

[Feature | Fix | Documentation | Other() ]

Fix

Short Description

1. Fixed max_tokens limit in src/utils/model_client_utils.py: reduced from 30,720 to 16,384 to match gpt-4o's limit, preventing API errors.
2. Fixed JSON parsing in src/utils/json_utils.py: rewrote fix_common_json_errors() to handle LaTeX backslashes ((, ), \to, \lim, etc.) by escaping them correctly, and added a fallback in the exception handler. This resolved JSON parsing errors in task generation and problem solving.

Tests Added

...

---

This change is 

[afkanpour]

@afkanpour reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @Ali-Meh619)

---

src/utils/json_utils.py line 0 at r1 (raw file):
Please add a unit test for the changes you make below. The test module should be located under https://github.com/VectorInstitute/automated_capability_evaluation/tree/main/tests/src

---

src/utils/json_utils.py line 51 at r1 (raw file):

    # Valid JSON escapes: \", \\, \/, \b, \f, \n, \r, \t, \uXXXX
    # LaTeX uses \(, \), \[, \], \epsilon, \delta, etc. which need to be escaped as \\(, \\), etc.
    

Please remove all instances of trailing whitespaces.

Code quote:

···