I am currently using the following method to restore from a checkpoint (check out this answer)
checkpoint = tf.train.Checkpoint(model=model)
checkpoint.read(os.path.join(
dir_path, model_name)).expect_partial()
However from what I can tell, neither Checkpoint.init nor read supports passing in a subset of variables to restore. It is a very common use case to restore some of the variables and re-initialize the others from scratch.
On the other hand, tf.train.Saver, which does support a variable list, seems to be for v1 non-eager mode, though correct me if I am mistaken.
