Problem

CI builds are failing due to 215 known test failures that are platform-specific and documented. Need to implement test exclusions to achieve green builds while maintaining 99%+ test coverage.

Objective

Configure the RHEL test workflow to exclude known failing tests, achieving green CI builds without losing test coverage.

Current Status

• Total tests: ~20,400
• Failing tests: ~215 (1.05%)
• Pass rate: ~99%
• All failures are documented and understood

Tests to Exclude

1. inductor/test_cutlass_backend - 189 tests (CUTLASS library not installed)
2. inductor/test_flex_attention - 24 tests (H200 float16 alignment)
3. test_cpp_extensions_jit - 1 test (cuDNN JIT compilation)
4. test_partial_flat_weights - 1 test (RNN parameter handling)

Expected Results

• All 5 test shards pass
• ~99% test coverage maintained
• Core PyTorch functionality fully tested
• CI builds show green status

Acceptance Criteria

• [ ] Test exclusions added to _linux-test.yml
• [ ] All exclusions documented with issue numbers
• [ ] CI pipeline runs successfully with green status
• [ ] Test coverage metrics confirm 99%+ coverage

References

• Current workflow: .github/workflows/_linux-test.yml
• Documentation: 5_SHARD_FAILURE_SUMMARY.md