resume train

[2016110071]

设置了resume_path之后，训练的时候

[rank0]: File "src/flow_factory/trainers/grpo.py", line 57, in init
[rank0]: super().init(**kwargs)
[rank0]: File "src/flow_factory/trainers/abc.py", line 65, in init
[rank0]: self._initialization()
[rank0]: File "src/flow_factory/trainers/abc.py", line 207, in _initialization
[rank0]: self.optimizer = self._init_optimizer()
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "src/flow_factory/trainers/abc.py", line 162, in _init_optimizer
[rank0]: self.optimizer = torch.optim.AdamW(
[rank0]: ^^^^^^^^^^^^^^^^^^
[rank0]: File "python3.12/site-packages/torch/optim/adamw.py", line 37, in init
[rank0]: super().init(
[rank0]: File "python3.12/site-packages/torch/optim/adam.py", line 101, in init
[rank0]: super().init(params, defaults)
[rank0]: File "python3.12/site-packages/torch/optim/optimizer.py", line 395, in init
[rank0]: raise ValueError("optimizer got an empty parameter list")
[rank0]: ValueError: optimizer got an empty parameter list

请问是为什么，应该怎么修改

[Jayce-Ping]

resume_path是什么呢？能否提供完整的config

[2016110071]

resume_path是什么呢？能否提供完整的config

resume_path: "save/v19/qwen-image-edit-plus_lora_grpo_20260324_194638/checkpoints/checkpoint-24/"
resume_type: 'lora'

[Jayce-Ping]

看起来没什么问题，请问是否启用了fsdp2或者fsdp_full_shard？