`compression=None` to `tf.data.experimental.service.distribute(...)`.
    *   `tf.data.Dataset.batch()` now supports `num_parallel_calls` and
        `deterministic` arguments. `num_parallel_calls` is used to indicate that
        multiple input batches should be computed in parallel. With
        `num_parallel_calls` set, `deterministic` is used to indicate that
        outputs can be obtained in the non-deterministic order.

max-width: percentage($size / $columns);\n}\n\n@mixin make-col-auto() {\n  flex: 0 0 auto;\n  width: auto;\n  max-width: 100%; // Reset earlier grid tiers\n}\n\n@mixin make-col-offset($size, $columns: $grid-columns) {\n  $num: $size / $columns;\n  margin-left: if($num == 0, 0, percentage($num));\n}\n\n// Row columns\n//\n// Specify on a parent element(e.g., .row) to force immediate children into NN\n// numberof columns. Supports wrapping to new lines, but does not do a Masonry\n// style grid.\n@mixin row-cols($count)...

max-width: percentage($size / $columns);\n}\n\n@mixin make-col-auto() {\n  flex: 0 0 auto;\n  width: auto;\n  max-width: 100%; // Reset earlier grid tiers\n}\n\n@mixin make-col-offset($size, $columns: $grid-columns) {\n  $num: $size / $columns;\n  margin-left: if($num == 0, 0, percentage($num));\n}\n\n// Row columns\n//\n// Specify on a parent element(e.g., .row) to force immediate children into NN\n// numberof columns. Supports wrapping to new lines, but does not do a Masonry\n// style grid.\n@mixin row-cols($count)...