{"id":"6f26adbc-8f70-4421-971f-8e63d9a97401","url":"https://api.tumn.com/news/6f26adbc-8f70-4421-971f-8e63d9a97401","title":"Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM","summary":"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 T","sections":[],"faq":[],"sources":[{"url":"https://aws.amazon.com/blogs/machine-learning/push-based-kv-transfer-to-improve-ttft-and-tpot-for-disaggregated-inference-in-vllm/","title":"AWS Machine Learning Blog","publishedAt":"2026-10-19T16:05:17.000Z"}],"author":"TUMN 큐레이션","aiAssisted":false,"reviewStatus":"automated","reviewNote":"원문 제목과 공개된 발췌를 인용하고 원문으로 연결합니다. 내용을 재작성하거나 요약하지 않았습니다.","publishedAt":"2026-10-19T16:05:17.000Z","modifiedAt":"2026-10-19T21:25:34.000Z","checkedAt":"2026-10-19T21:25:34.296Z","language":"ko","kind":"link","sourceLabel":"AI 공식","profile":null,"via":null,"formats":{"json":"https://api.tumn.com/news/6f26adbc-8f70-4421-971f-8e63d9a97401.json","markdown":"https://api.tumn.com/news/6f26adbc-8f70-4421-971f-8e63d9a97401.md"}}