NotImplementedError: must be implemented in descendants on train_loop method in object_detection.model_lib_v2

NotImplementedError Traceback (most recent call last)
in ()
----> 1 train_loop(pipeline_config,model_dir,train_steps=10,checkpoint_every_n=100,checkpoint_max_to_keep=5,save_final_config=True)

2 frames
/usr/local/lib/python3.6/dist-packages/object_detection/model_lib_v2.py in train_loop(pipeline_config_path, model_dir, config_override, train_steps, use_tpu, save_final_config, checkpoint_every_n, checkpoint_max_to_keep, **kwargs)
530 # is the chief.
531 summary_writer_filepath = get_filepath(strategy,
--> 532 os.path.join(model_dir, 'train'))
533 summary_writer = tf.compat.v2.summary.create_file_writer(
534 summary_writer_filepath)

/usr/local/lib/python3.6/dist-packages/object_detection/model_lib_v2.py in get_filepath(strategy, filepath)
385 for the chief.
386 """
--> 387 if strategy.extended.should_checkpoint:
388 return filepath
389 else:

/usr/local/lib/python3.6/dist-packages/tensorflow/python/distribute/distribute_lib.py in should_checkpoint(self)
2325 def should_checkpoint(self):
2326 """Whether checkpointing is needed."""
-> 2327 raise NotImplementedError("must be implemented in descendants")
2328
2329 @Property

NotImplementedError: must be implemented in descendants

I am using the model SSD ResNet152 V1 FPN 640x640 from Tensorflow model zoo
Corresponding config file is being attached
config.txt

And I am using the direct method train_loop from object_detection.model_lib_v2

For conversion of Pascal VOC to tfrecord fromat is done using create_pascal_tf_record

I didn't understand the issue, any suggestions are welcome.
Thanks in advance