alibaba / alibaba/ROLL

ifeval_rule_reward_worker bug

Open
#125 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
3.4k
Forks
312
Avg merge
1h 2m
Merged PRs (30d)
2

Description

1. else 后还要加个 `result = False`,否则 result 可能未定义,导致后面抛异常

https://github.com/alibaba/ROLL/blob/445a029c5c2cc75cf0dc896b3c5bcd72eda1ee2b/roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py#L635

2. 这里 continue 之前需要加个 `repetition_penalty_rewards.append(0.0)`,否则 repetition_penalty_rewards 维度和 results 不一致,导致后面相加时会变成短的,转tensor时抛异常

https://github.com/alibaba/ROLL/blob/445a029c5c2cc75cf0dc896b3c5bcd72eda1ee2b/roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py#L621

Contributor guide

No contributing guide indexed for this repository

Research direction

Open roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py around lines 621 and 635 and inspect the branches described in the issue. Verify that the result variable is defined after the else branch and that repetition_penalty_rewards stays aligned when continue is reached, then run the relevant available checks to confirm tensor conversion no longer fails.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
55/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.