diff --git a/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py b/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py index f518076c7..0e67efcf9 100755 --- a/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py +++ b/applications/DeepSpeed-Chat/dschat/rlhf/rlhf_engine.py @@ -269,8 +269,8 @@ def _init_reward(self, critic_model_name_or_path): zero_stage = 0 ds_config = get_eval_ds_config(offload=self.args.offload_reward_model, - dtype=self.args.dtype, - stage=zero_stage) + dtype=self.args.dtype, + stage=zero_stage) # We need to set train batch size and micro batch size here to pass the sanity check of DeepSpeed engine. ds_config[