Objective

Set up test infrastructure using self-hosted runner with H200 GPU.

Work Completed

• Configured self-hosted runner: test-runner-git-109-vpc
• Set up Podman container runtime (Docker emulation)
• Configured GPU passthrough for CUDA tests
• Implemented test sharding (initially 3 shards, optimized to 5)
• Added test artifact collection and reporting
• Set up --keep-going flag for comprehensive coverage

Test Execution

• Full PyTorch test suite: ~20,400 tests
• Test shards: 5 (optimized from 3)
• Parallel execution within shards
• Comprehensive logging and artifact collection

Deliverables

• [x] Self-hosted runner operational
• [x] GPU passthrough working
• [x] Test execution reliable
• [x] Artifact collection functional

References

• Workflow: .github/workflows/rhel-build-test.yml
• Test script: .ci/pytorch/test.sh