Andres Lugo

Hi @zou3519 This PR addresses several ROCm only functorch failures. Many of which are referenced in this issue: https://github.com/pytorch/pytorch/issues/96560 Will push up a follow up PR addressing the remaining failing...

@pytorchbot label ciflow/trunk

> Hi @alugorey, thanks for fixing this. It seems there are two separate issues, one is forward AD and the other is the contiguous tensor assertion. Is it possible to...

@andrewor14 Actually, on closer inspection the reason i wrapped up both those issues into one is because they both fell under the umbrella of the same skip decorator in test_ops.py....

> > @andrewor14 Actually, on closer inspection the reason i wrapped up both those issues into one is because they both fell under the umbrella of the same skip decorator...

@xinyazhang Could you please add a comment to this PR showing test_linalg.py TestLinalgCUDA.test_svd_* passes for each data type? Make sure to force the `gesvda` path to verify the new gesvda...

> Hi @alugorey do you know how to enable the logging of rocSOLVER? I've tried `ROCSOLVER_LAYER=7 ROCSOLVER_LEVELS=99 ROCSOLVER_LOG_TRACE_PATH=t ROCSOLVER_LOG_BENCH_PATH=b ROCSOLVER_LOG_PROFILE_PATH=p` but it doesn't work hi @xinyazhang , sorry i'm just...

I need to do a rebase

This code was brought in as a part of these 2 commits https://github.com/ROCm/pytorch/commit/69c6e0b8516933d08ad54bb2756124d98300cc5b https://github.com/ROCm/pytorch/commit/2cb3301f807e2d58d719bbd976b68a13999f1c70 They have already been merged into rocm6.2_internal_testing. We can abandon this PR

@jithunnair-amd @bmedishe Jithun is correct. This error is a direct side effect of the lack of weight flattening in our version of RNNs.