[None][feat] Optimize CUDA graph memory usage for spec decode cases#6718

mikeiovine merged 1 commit intoNVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

mikeiovine:cuda-graph-mem-savingmikeiovine/TensorRT-LLM:cuda-graph-mem-savingCopy head branch name to clipboard