Then you can reload it as shown below: from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint checkpoint_dir = os.path.join(trainer.args.output_dir, "checkpoint-final") trainer.deepspeed.save_checkpoint(checkpoint_dir) fp32_model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir) Once load_state_dict_from_zero_checkpoint is run, the model is no longer usable in DeepSpeed in the context of the same application.