add ppo model

forward_demo.py

@@ -263,6 +263,8 @@ if __name__ == "__main__":
         trainer.strategy.config["zero_optimization"]["allgather_bucket_size"] = args.ds_bucket_mb * 1000 * 1000
         trainer.strategy.config["zero_optimization"]["reduce_bucket_size"] = args.ds_bucket_mb * 1000 * 1000
 
+    model.deepspeed_offload()
+
     seq = torch.randint(0, 50277, (1, 100))
     model(seq)