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

@ -80,7 +80,7 @@ class PoolMatrixDataset(ProceduralDataset):
def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
"""Score the answer based on the metadata"""
if not answer:
if not isinstance(answer, str):
return 0.0
reward = 0.0
@ -91,8 +91,6 @@ class PoolMatrixDataset(ProceduralDataset):
reward = 1.0
elif oracle_answer.shape == answer.shape:
reward = 0.1
else:
reward = 0.01
except Exception:
pass
return reward