1  ---
2  enhancements:
3    - |
4      Use `device_map` when loading a TransformersSimilarityRanker and ExtractiveReader.
5      This allows for multi-device inference and for loading quantized models (e.g. load_in_8bit=True)