NGC PyTorch Container Pytorch Versions

Looking at the support matrix for NGC containers Frameworks Support Matrix :: NVIDIA Deep Learning Frameworks Documentation, all of the images that contain PyTorch use alpha builds from particular git revisions rather than from PyTorch release versions. I was just curious what the motivation for that is. From a development standpoint I would like to be able to use the containers as a basis for my development, but then I want to be able to use the same version as a dependency where my models will actually run. That said, it feels a little bit awkward to use non release builds unless there is a particular reason not to.

Thanks in advance.