search

ROCm / pytorch

Tensors and Dynamic neural networks in Python with strong GPU acceleration
http://pytorch.org
Other
219 stars 50 forks source link

[release/2.2] fix test_oom_tracing and skip test_profiler_ cuda_sync_events #1451

Closed dnikolaev-amd closed 1 week ago

dnikolaev-amd commented 1 week ago

test_oom_tracing expects OutOfMemory exception by allocating a large tensor. MI300X has enough memory to allocate test tensor This PR increases tensor size with a large margin to force OutOfMemory exception on MI300X and future GPU generations

@SkipIfRocm for test_profiler_cuda_sync_events was cherry-picked from upstream https://github.com/ROCm/pytorch/commit/57fb8860f3d55c2f24986e4afbd643766c649a03 for SWDEV-452384