Modified logging threshold for memory profiling (#102243)
Fixed test_memory_profiler::TestMemoryProfilerE2E::test_memory_timeline by changing the (arbitrary) threshold for logging. We observe differently-sized allocations on different AMD GPUs, so chose a higher threshold over 512 to account for those differences and yet satisfy the test requirements.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102243
Approved by: https://github.com/ezyang
