This paper received very good scores, with strong consensus in the reviews. It presents interesting ideas for doing lossless compression with flow-based models. The paper could be strengthened with more experiments/comparisons, but it still a good contribution to Neurips.