You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We'll do a git filter-repo on nvfuser folder and strip pytorch core code-base from nvfuser.
I tried it with current nvfuser devel branch + a pytorch master repo from upstream. Currently there's some expected failures:
dynamo + primTorch: Since we updated our python APIs, and will continue to do so. Related changes needs to be pushed to upstream. But we will need to figure out how we would want to do that. I think the easier way is that we do the filter-repo first on our branch (basically ignoring those failures) and once we have that, we'll be able to convert upstream pytorch third_party/nvfuser into using a submodule instead of real code there. Need to follow this up with upstream folks.
TorchScript test in test_jit_cuda_fuser.py. There were some tests related to gather and index_select that are failing. For gather, I know it's coming from our changes in symbolic_scripts, which is not pushed to upstream. We'll need to do that. Not any clue on why index_select is failing though. Would need some investigation.
Alternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered:
馃殌 The feature, motivation and pitch
We'll do a git filter-repo on nvfuser folder and strip pytorch core code-base from nvfuser.
I tried it with current nvfuser devel branch + a pytorch master repo from upstream. Currently there's some expected failures:
dynamo + primTorch: Since we updated our python APIs, and will continue to do so. Related changes needs to be pushed to upstream. But we will need to figure out how we would want to do that. I think the easier way is that we do the filter-repo first on our branch (basically ignoring those failures) and once we have that, we'll be able to convert upstream pytorch
third_party/nvfuser
into using a submodule instead of real code there. Need to follow this up with upstream folks.TorchScript test in
test_jit_cuda_fuser.py
. There were some tests related togather
andindex_select
that are failing. For gather, I know it's coming from our changes in symbolic_scripts, which is not pushed to upstream. We'll need to do that. Not any clue on why index_select is failing though. Would need some investigation.Alternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered: