Code update request

[Jiangchenglin521]

Hello
I'd like to ask if it's possible to update the initialization code of the trainer in dpo.py. Since it's already incompatible with the trl code after version 1.00, it's somewhat outdated. This can greatly save everyone's time spent on debugging.

[WeiXiongUST]

Thanks for bringing this to us. We will adapt the code accordingly. For now you may try use older trl. The try code update is mainly for the deep RL methods recently and DPO code is quite stable since then.