NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-19 12:57:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

re-append stop in math training path

Browse source
This commit is contained in:
Partho Das 2026-02-23 22:50:38 +05:30
parent f02c24204d
commit adf075112c
1 changed files with 4 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
environments/math_server_zero.py
View file

@ -421,6 +421,10 @@ class MathEnv(BaseEnv):
if self.config.mask_too_long_completions:
scores["overrides"][-1]["set_advantage_to_zero"] = True
else:
# re-append </answer> if stripped by vLLM stop string handling
# (mirrors the eval path in rollout_and_score_eval)
if ("<answer>" in resp) and ("</answer>" not in resp):
resp = resp + "</answer>"
task = loop.run_in_executor(self.mp_executor, score_answer, gold, resp)
reward = await task
if reward is None: