pytorch

25725fd6 - (Re-open) Adds cudaMallocAsync as an alternative backend for the CUDA allocator (#82682)

(Re-open) Adds cudaMallocAsync as an alternative backend for the CUDA allocator (#82682) Rebased version of @mcarilli 's cudaMallocAsync #65365 for continued testing Pull Request resolved: https://github.com/pytorch/pytorch/pull/82682 Approved by: https://github.com/ngimel