Lilt continually invests in its systems design and deployment configurations to keep its machine translation fast and highly available. Here, we describe how we re-structured the code used to translate segments in batches in order to better parallelize the GPU resources available.
