Proportion of inference volume run asynchronously at batch rather than interactive rates.
Batch-processed calls ÷ total calls × 100