[None][perf] One-model spec decode: vocab-parallel token selection, fused DSpark Markov chain and spec tail, drafter TP fusion - #20940

[None][perf] One-model spec decode: vocab-parallel token selection, fused DSpark Markov chain and spec tail, drafter TP fusion#20940

longlee0622 merged 5 commits into

NVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

longlee0622:user/jonasl/k3-pr-speclonglee0622/TensorRT-LLM:user/jonasl/k3-pr-specCopy head branch name to clipboard