[jamesblonde]

You could use Tensorflow-on-Spark to read your JSON into a RDD in Spark. Then the Tf-RDD-Reader will be in-memory and can feed your training.