FUJIsyu0515
Results
3
comments of
FUJIsyu0515
@yuhangzang @cool-xuan The methods you provided are very useful for avoiding OOM at startup. I have tried them. However, now it always suddenly appears OOM after running dozens of steps....
@czczup 请问这两个数据集最近有计划传完吗？
just use open-rlhf or verl to train with DAPO or GSPO, it's better to use frameworks that supporting distributed training for rl.