Jack Taylor
Jack Taylor
jenkins retest this please
jenkins retest this please
jenkins retest this please
@antiagainst @jansel cc: @jeffdaily @jithunnair-amd Looks like we are seeing slight inaccuracy with ROCm causing this UT to fail. I took a deeper look, with a standalone comparison fp16/fp32 models...
There's a few cpu failures for the conv3d test, maybe conv3d cpu doesn't support max autotune cc: @jansel Will take a look. Rebasing to check if `inductor/test_compiled_autograd.py::TestCompiledAutograd::test_free_activation_memory_subclass` failure is flakey
@pytorchbot rebase
Rebasing again due to minifier issues
@pytorchbot merge
Hi @jansel do we need to do more than skip on ROCm to make sure we don't see issues in other workloads? (i.e. do we need to add a way...
> @pragupta , The changes look good. Can we merge this after the upstream patch is merged. cc: @pruthvistony not sure if this upstreamed yet, but we need this for...