AI · Research · 2026-11-20 09:02
WavePP introduces a pipeline parallelism approach to optimize the prefill phase of large language models under prefix reuse scenarios, significantly improving system throughput.
Auto-generated summary for reference only. RSS