Why is the reward is equal to 1 if the gold solution is not parseable? Shouldn't it be 0?

[asirgogogo]

[ali-cogni]

If the gold solution cannot be parsed, then assigning a reward of 0 or 1 has the same effect. The advantage is based on
(𝑟−mean(r))/std(𝑟).

If all rewards are 0 or 1, the advantage will be 0 in both cases.
More discussion here:

#144 (comment)

[asirgogogo]

I get it, thank you！！！

[asirgogogo]

@ali-cogni
Can you provide a available docker image? I am deeply troubled by the environment.