[PyTorch] Share CUDA graph memory across dynamic CP variants - #3353

[PyTorch] Share CUDA graph memory across dynamic CP variants#3353

xiaoyao0115 wants to merge 3 commits into

NVIDIA:release_v2.14NVIDIA/TransformerEngine:release_v2.14from

xiaoyao0115:test/dcp-cg-gap-fixxiaoyao0115/TransformerEngine:test/dcp-cg-gap-fixCopy head branch name to clipboard