Transforming tensorflow v1 graph and weights into saved model

I defined model (mnist digits recognition) using tensorflow 2.15.0 and tensorflow.compat.v1. Model was **not ** trained and the graph was exported using following code:

init = tf.global_variables_initializer()
saver_def = tf.train.Saver().as_saver_def()

with open('frozen_models/graph_v1.pb', 'wb') as f:
  f.write(tf.get_default_graph().as_graph_def().SerializeToString())

Later I loaded this graph into C++ application and trained model using tensorflow c_api (version 2.18.0). I saved checkpoint using operation defined by python call to tf.train.Saver().as_saver_def(). The output files of this operation are:

• checkpoint_1.data-00000-of-00001
• checkpoint_1.index

The code was basically something like that

// initialization

save_op = TF_GraphOperationByName(graph, "save/control_dependency");
checkpoint_file.oper = TF_GraphOperationByName(graph, "save/Const");
checkpoint_file.index = 0;

// later

TF_Output inputs[] = { checkpoint_file };
TF_Tensor* input_values[] = { tensor };

const TF_Operation* ops[] = {
    type == CheckpointType::Save ? save_op : restore_op
};

TF_SessionRun(
    session,
    NULL,
    /* Inputs */
    inputs,
    input_values,
    1,
    /* Outputs */
    NULL,
    NULL,
    0,
    /* Init operation */
    ops,
    1,
    NULL,
    status
);

My question is - how can I load this graph into tensorflow (preferably 2.x.x) via python and restore weights so the model could be exported in SavedModel format? Is this even possible?

Disclaimer - I know that it would be easier to just to it in python but my goal is to learn how that toolchain/formats work, not to create ML model.

I tried few approaches but maybe I'm missing some necessary piece.

For example I imported graph using following code, but I don't really know where to go from here:

import tensorflow.compat.v1 as tf

with tf.io.gfile.GFile(graph_filename, "rb") as f:
    graph_def = tf.get_default_graph().as_graph_def()
    graph_str = f.read()
    graph_def.ParseFromString(f.read())