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 on Amazon EC2 P5en instances with AWS EFA.
Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM
Publication summary. The terms of AWS Machine Learning do not allow reproduction of the full article: read the complete text on the original website.
Source summary (English)
Read the full article
aws.amazon.com