Migrate HuggingFace Tests to Granite 4 When Adapters Available
Background
During the Granite 3 → 4 migration (#344), all backends and tests were successfully migrated to Granite 4 hybrid models except HuggingFace tests, which require aLoRA adapters that don't yet exist for Granite 4.
Current State
- HuggingFace tests: Use
granite-3.3-8b-instruct (hardcoded in test/backends/test_huggingface.py)
- All other backends: Successfully migrated to Granite 4 hybrid models
- Blocking issue: Missing
requirement_check adapter for Granite 4
What's Missing
HuggingFace tests use two adapters:
- ✅
answerability - Available for Granite 4.0-micro in ibm-granite/granite-lib-rag-r1.0
- ❌
requirement_check - Only available for Granite 3.x in ibm-granite/rag-intrinsics-lib
What Needs to Happen
Once Granite 4 adapters for requirement_check are released, update test/backends/test_huggingface.py to use Granite 4 model and test. Consider using smaller model (granite-4.0-micro) to reduce memory requirements.
References
Migrate HuggingFace Tests to Granite 4 When Adapters Available
Background
During the Granite 3 → 4 migration (#344), all backends and tests were successfully migrated to Granite 4 hybrid models except HuggingFace tests, which require aLoRA adapters that don't yet exist for Granite 4.
Current State
granite-3.3-8b-instruct(hardcoded intest/backends/test_huggingface.py)requirement_checkadapter for Granite 4What's Missing
HuggingFace tests use two adapters:
answerability- Available for Granite 4.0-micro inibm-granite/granite-lib-rag-r1.0requirement_check- Only available for Granite 3.x inibm-granite/rag-intrinsics-libWhat Needs to Happen
Once Granite 4 adapters for
requirement_checkare released, updatetest/backends/test_huggingface.pyto use Granite 4 model and test. Consider using smaller model (granite-4.0-micro) to reduce memory requirements.References