Thomas Viehmann
Thomas Viehmann
Yeah, I'll learn to read...
@jjsjann123 what's your view on this? The CI seemed to pass earlier, but the PR is still a draft and thus not merged...
To facilitate discussion, I generated these numbers by adding `thunder/tests/conftest.py`. If raising an error there leads to the test failing, we could use a similar setup for enforcing a GPU...
The numbers are GB GPU mem. We can run tests needing 8GB, but we should not run them in the parallel setup. For comparison in #394 I listed all tests...
This is the oldest non-draft PR with all green CI. Do we want this as is or do we want to change things yet, @crcrpar @IvanYashchuk ?
I think it is orthogonal.
Not sure what I can say beyond the linked #392. These tests use multiple GB of GPU memory leading to OOM when run with the normal tests. If reducing the...
@IvanYashchuk I would propose to close this for now, as it has conflicts and the actual solution is move involved, as in it would likely not be easier to update...
Thank you! So one prominent pattern is `make_callable_legacy(vjp(fn))`. I could be wrong, but I think that this will likely need work signature changing transforms / updating of prologues to make...
We do have a functionality gap for signature-changing transforms as they need adapting of the prologue.