jax
eed75d72
- [ROCm] Run export_harnesses_multi_platform_test_gpu downstream
Go
Login via GitHub
Home
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
45 days ago
[ROCm] Run export_harnesses_multi_platform_test_gpu downstream The fork is where we want the wider coverage. Observed flake rate on the gfx950 RBE pool is 1 in 102 genuine executions over five weeks.
References
#890 - fix: reduce DPX workers_per_gpu 2→1 to avoid HBM exhaustion
Author
magaonka-amd
Committer
magaonka-amd
Parents
8eb6aff6
Loading