[BUG] restart file from v2.0.0.b4 cannot freeze models in v2.0.0

[njzjz]

Summary

Continue training from a checkpoint file in v2.0.0.b4 cannot freeze models in v2.0.0. #1000 breaks it. It worked after I removed model_type, train_attr/min_nbor_dist, and train_attr/training_script.

The following nodes will be frozen: ['model_type', 'descrpt_attr/rcut', 'descrpt_attr/ntypes', 'model_attr/tmap', 'model_attr/model_type', 'model_attr/model_version', 'train_attr/min_nbor_dist', 'train_attr/training_script', 'o_energy', 'o_force', 'o_virial', 'o_atom_energy', 'o_atom_virial', 'fitting_attr/dfparam', 'fitting_attr/daparam']
Traceback (most recent call last):
  File "/home/jz748/ambertools-dp/bin/dp", line 10, in <module>
    sys.exit(main())
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/deepmd/entrypoints/main.py", line 439, in main
    freeze(**dict_args)
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/deepmd/entrypoints/freeze.py", line 183, in freeze
    output_graph_def = tf.graph_util.convert_variables_to_constants(
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/util/deprecation.py", line 337, in new_func
    return func(*args, **kwargs)
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/framework/graph_util_impl.py", line 276, in convert_variables_to_constants
    ret = convert_to_constants.convert_variables_to_constants_from_session_graph(
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/framework/convert_to_constants.py", line 1160, in convert_variables_to_constants_from_session_graph
    converter_data=_SessionConverterData(
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/framework/convert_to_constants.py", line 857, in __init__
    graph_def = graph_util.extract_sub_graph(graph_def, output_node_names)
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/util/deprecation.py", line 337, in new_func
    return func(*args, **kwargs)
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/framework/graph_util_impl.py", line 209, in extract_sub_graph
    _assert_nodes_are_present(name_to_node, dest_nodes)
  File "/home/jz748/ambertools-dp/lib/python3.9/site-packages/tensorflow/python/framework/graph_util_impl.py", line 164, in _assert_nodes_are_present
    assert d in name_to_node, "%s is not in graph" % d
AssertionError: model_type is not in graph

Deepmd-kit version, installation way, input file, running commands, error log, etc.

v2.0.0

Steps to Reproduce

Train a model in v2.0.0.b4 without freezing, and freeze in v2.0.0.

Further Information, Files, and Links

[amcadmus]

This is expected, because to support compressed training, 2.0.0 breaks the model compatibility of 2.0.0.b4.

[amcadmus]

we shouldn't have introduced such a change in beta version...

[njzjz]

I don't find a easy way to provide compatibility..