System information

• TensorFlow version (you are using): master (Nov. 30, 2018)
• Are you willing to contribute it (Yes/No): Yes

Describe the feature and the current behavior/state.
When using a DistributionStrategy with a Keras model that includes sample_weights, a 'NotImplemented' exception is thrown. For example,

   strategy = mirrored_strategy.MirroredStrategy()
    model.compile(loss='categorical_crossentropy',
                    optimizer=opt,
                    metrics=['accuracy'],
                    weighted_metrics=['accuracy'],
                    sample_weight_mode='temporal',
                    distribute=strategy)

The exception originates in tensorflow/python/keras/engine/training.py

 if sample_weight_mode:
	        raise NotImplementedError('sample_weight_mode is not supported with '
	                                  'DistributionStrategy.')

Can this be supported or can you give an estimated timeline?

Many thanks

Will this change the current api? How?
No. It looks like the API should support this.

Who will benefit with this feature?
Anyone using keras models with sample_weights in a DistributedStrategy mode.

Any Other info.