Entity

Disaggregated serving

An architectural approach that separates prefill (initial prompt processing) and decode (token generation) operations across different compute resources to optimize throughput and latency trade-offs.