In order to facilitate pre-computing the next DAG, we should have a function which amounts to a "Generate & Save" that doesn't require the entire DAG in memory - so the current epoch DAG can be kept in memory, and the next epoch can be generated with low memory
In order to facilitate pre-computing the next DAG, we should have a function which amounts to a "Generate & Save" that doesn't require the entire DAG in memory - so the current epoch DAG can be kept in memory, and the next epoch can be generated with low memory