Fine-tune 之后，使用预测后的模型报错

[wb-hwang]

欢迎您反馈PaddleHub使用问题，非常感谢您对PaddleHub的贡献！
在留下您的问题时，辛苦您同步提供如下信息：

• 版本、环境信息
  1）PaddleHub和PaddlePaddle版本：PaddlePaddle 2.4.0，paddleHub 2.3.0
  2）系统环境：AIStudio，python 1.7.4，
• 复现信息：如为报错，请给出复现环境、复现步骤
  直接运行教程里面迁移学习-图像分类 的代码，完成迁移学习后，加载模型报错。
  使用AIstudio、本地 运行都出现一样的错误。

迁移学习成功：

[2023-07-10 11:45:44,857] [    EVAL] - Saving best model to img_classification_ckpt/best_model [best acc=0.6710]
[2023-07-10 11:45:44,860] [    INFO] - Saving model checkpoint to img_classification_ckpt/epoch_1

加载模型的代码：

import paddle
import paddlehub as hub

if __name__ == '__main__':

    model = hub.Module(name='resnet50_vd_imagenet_ssld', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='img_classification_ckpt/')
    result = model.predict(['flower.jpg']) 

报错：

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/setuptools/depends.py:2: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
Traceback (most recent call last):
  File "untitled.py", line 6, in <module>
    model = hub.Module(name='resnet50_vd_imagenet_ssld', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='img_classification_ckpt/')
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddlehub/module/module.py", line 396, in __new__
    **kwargs)
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddlehub/module/module.py", line 511, in init_with_name
    return user_module_cls(**kwargs)
  File "/home/aistudio/.paddlehub/modules/resnet50_vd_imagenet_ssld/module.py", line 240, in __init__
    self.model_dict = paddle.load(load_checkpoint)
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/framework/io.py", line 1103, in load
    load_result = _legacy_load(path, **configs)
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/framework/io.py", line 1150, in _legacy_load
    load_result = _load_state_dict_from_save_params(model_path)
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/framework/io.py", line 147, in _load_state_dict_from_save_params
    attrs={'file_path': os.path.join(model_path, name)},
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/dygraph/tracer.py", line 314, in trace_op
    stop_gradient, inplace_map)
  File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/dygraph/tracer.py", line 176, in eager_legacy_trace_op
    returns = function_ptr(*arg_list, *attrs_list)
ValueError: (InvalidArgument) Deserialize to tensor failed, maybe the loaded file is not a paddle model(expected file format: 0, but 1667433344 found).
  [Hint: Expected version == 0U, but received version:1667433344 != 0U:0.] (at /paddle/paddle/phi/core/serialization.cc:106)
  [operator < load > error]

迁移学习后，生成的文件结构是：

-img_classification_ckpy/
  -- best_model/
    --- model.pdopt
    --- model.pdparams
  -- epoch_1/
  -- visualization/
  -- metrics.pkl

做过以下尝试，依旧不行

1. 切换paddle-gpu、paddle-cpu
2. AIstudio、本地（windows 11)跑
3. 切换paddle 版本到2.3.0