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

@ -92,14 +92,14 @@ class PalindromeDataset(ProceduralDataset):
- Correct answer (palindrome with only correct letters in the correct quantities) gives 1.0
- An answer that is a palindrome, but not with the same letters as provided, gives 0.05
- An answer that is a string, but not a palindrome gives 0.02
- An empty string gives 0.01.
- An empty string gives 0.0
- None gives 0.0.
"""
if answer is None or not isinstance(answer, str):
return 0.0 # No answer given
if answer == "":
return 0.01
return 0.0
metadata = entry["metadata"]
answer = answer.strip().lower()