Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM
vLLM's prefill-decode disaggregation splits LLM inference across engine instances but must transfer the KV cache between them. This post introduces a push-based NIXL connector that moves transfer coordination off the critical path, delivering up to 3x better TTFT and 30% better TPOT in benchmarks o…
Read the full story at AWS Machine Learning Blog ↗
Timeline · 1 report
- 2026-10-19 16:05 · AWS Machine Learning Blog
Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM