Use runs-on GPU runners for CI#1439

Merged

dfalbel merged 10 commits into

mainfrom

Jul 16, 2026

Commits on Apr 28, 2026

Use runs-on GPU runners instead of self-hosted for GPU CI jobs
dfalbel
committed
Serialize GPU jobs to respect single-instance AWS quota
dfalbel
committed
Add concurrency group to GPU jobs to prevent parallel instance creation
dfalbel
committed
Add skip_slow_tests() helper and skip cuda_memory_snapshot in cudatoolkit job
dfalbel
committed
Mark autocast grad scaler training loop test as slow
dfalbel
committed
Disable cuda_record_memory_history after snapshot test via withr::defer
dfalbel
committed
Add retry=false to GPU runner labels to always use spot instances
dfalbel
committed
Explicitly set spot=true on GPU runner labels
dfalbel
committed
Keep spot=true but remove retry=false to fix on-demand fallback issue
dfalbel
committed

Commits on Jul 16, 2026

Skip slow tests in GPU CI job
dfalbel
committed