opt reward model

src/rlhf/reward.py

@@ -136,7 +136,7 @@ class RewardModel(pl.LightningModule):
         strategy = self.trainer.strategy
         if isinstance(strategy, DeepSpeedStrategy):
             cfg = strategy.config["zero_optimization"]
-            return cfg.get("offload_optimizer") or cfg.get("offload_param")
+            return bool(cfg.get("offload_optimizer") or cfg.get("offload_param"))
         return False
 
     def single_forward(