NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-24 17:04:55 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

rm hardcoded same score check

Browse source
This commit is contained in:
Partho Das 2026-02-23 15:03:17 +05:30
parent 708b42a00f
commit f02c24204d
22 changed files with 0 additions and 85 deletions
Download patch file Download diff file Expand all files Collapse all files

8
environments/answer_format_environment/answer_format_environment.py
View file

@ -3653,14 +3653,6 @@ class AnswerFormatEnv(BaseEnv):
)
await self._save_failed_rollouts_to_jsonl()
# Check if all scores are the same (no learning signal)
if all(group_scores[0] == score for score in group_scores):
if self.debug_logging:
self.logger.debug(
"All scores are identical, returning None for learning signal"
)
return None
# Track successful groups for equivalent ratio enforcement
if self.ensure_equivalent_ratios:
# Count this as a successful group if we have any successful examples