Summary
Two SVD interpreter compatibility tests skip themselves at runtime due to fixture-config issues, not feature gaps.
Affected tests
tests/unit/model_bridge/compatibility/test_svd_interpreter.py:78: test_svd_interpreter_returns_different_answers_for_different_layers Skips with "Model only has one layer" (and yet the assertion is "different layers should differ")tests/unit/model_bridge/compatibility/test_svd_interpreter.py:113: test_svd_interpreter_returns_different_answers_for_different_models Skips with "Same model used for both fixtures" (and yet the assertion is "different models should differ")
Acceptance criteria
Where to start
- Inspect the
model / second_model fixtures (file-scoped or conftest.py), they're probably loading the same tiny model twice.
- Pick a tiny multi-layer model (e.g.,
attn-only-2l, solu-2l) for the multi-layer test and two distinct tiny models (e.g., solu-1l, attn-only-1l) for the cross-model test.
Summary
Two SVD interpreter compatibility tests skip themselves at runtime due to fixture-config issues, not feature gaps.
Affected tests
tests/unit/model_bridge/compatibility/test_svd_interpreter.py:78:test_svd_interpreter_returns_different_answers_for_different_layersSkips with"Model only has one layer"(and yet the assertion is "different layers should differ")tests/unit/model_bridge/compatibility/test_svd_interpreter.py:113:test_svd_interpreter_returns_different_answers_for_different_modelsSkips with"Same model used for both fixtures"(and yet the assertion is "different models should differ")Acceptance criteria
pytest.skip(...)is removed in both tests.Where to start
model/second_modelfixtures (file-scoped orconftest.py), they're probably loading the same tiny model twice.attn-only-2l,solu-2l) for the multi-layer test and two distinct tiny models (e.g.,solu-1l,attn-only-1l) for the cross-model test.