open-thought/reasoning-gym
1
0
Fork 0
mirror of https://github.com/open-thought/reasoning-gym.git synced 2026-04-23 16:55:05 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Minor question template & score_answer improvements (#261)

Browse source 
* 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
Download patch file Download diff file Expand all files Collapse all files

12
reasoning_gym/algorithmic/word_ladder.py
View file

@ -221,8 +221,8 @@ class WordLadderDataset(ProceduralDataset):
}
def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
if answer is None:
return 0
if not isinstance(answer, str):
return 0.0
answer_words = tuple(s.strip() for s in answer.upper().split(","))
@ -239,17 +239,17 @@ class WordLadderDataset(ProceduralDataset):
# 4. all words are in our vocabulary
if len(answer_words) < 2:
return 0
return 0.0
if answer_words[0] != start_word or answer_words[-1] != end_word:
return 0.01
return 0.0
if not all(len(w) == word_length for w in answer_words):
return 0.01
return 0.0
for i in range(1, len(answer_words)):
if sum(1 for a, b in zip(answer_words[i - 1], answer_words[i]) if a != b) != 1:
return 0.01
return 0.0
reward = 1.0
for word in answer_words: