When the installed PyTorch version is not nightly:
Issue an warning if torch.compile is requested.
Raise an error if both tensor-parallel and torch.compile are requested (tensor-parallel + torch.compile doesn't work on 2.1.2+cu121 due to a bug in functional collective. See https://github.com/pytorch-labs/gpt-fast/issues/64)
Stack from ghstack (oldest at bottom):
66
When the installed PyTorch version is not nightly:
torch.compile
is requested.torch.compile
are requested (tensor-parallel +torch.compile
doesn't work on2.1.2+cu121
due to a bug in functional collective. See https://github.com/pytorch-labs/gpt-fast/issues/64)