Running prefill and decode on separate GPU pools so each scales independently for better efficiency.
Related terms: Decode · GPU · Prefill
← All terms