Minor question template & score_answer improvements (#261)

* math prompt improvements
* ignore brackets in complex_arithmetic results
* improve additional instruction in prompt of polynomial_equations
* more strict tests for score_answer in polynomial_equations
* simplify special reward handling
* fix test_intermediate_integration
* fix sokoban dataset
* add common dataset score_answer consistency test
This commit is contained in:
Andreas Köpf 2025-03-04 21:55:09 +01:00 committed by GitHub
parent 061282e373
commit 5d7fbac0ad
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
106 changed files with 403 additions and 507 deletions

View file

@ -125,8 +125,8 @@ class ShortestPathDataset(ProceduralDataset):
def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
"""Overwrite this method in derived classes if a single oracle answer is not available."""
oracle_answer = entry["answer"].strip()
if answer is not None and len(answer) > 0:
if isinstance(answer, str) and len(answer) > 0:
oracle_answer = entry["answer"].strip()
answer = answer.strip()
# Exact answer
@ -145,8 +145,6 @@ class ShortestPathDataset(ProceduralDataset):
elif self._is_valid_path(matrix, answer):
return 0.5
return 0.01
return 0.0
def __getitem__(self, idx: int) -> dict: