[Refactor][Perf] Qwen3-omni: code predictor with re-prefill + SDPA and eliminate decode hot-path CPU round-trips#1758

hsliuustc0106 merged 10 commits into

vllm-project:mainvllm-project/vllm-omni:mainfrom

LJH-LBJ:qwen3-omni-decode-performanceLJH-LBJ/vllm-omni:qwen3-omni-decode-performanceCopy head branch name to clipboard