The community member is asking if anyone has had success using local Hugging Face models with RAG (Retrieval-Augmented Generation) evaluators for faithfulness and correctness. They would like to use the models "out-of-the-box" to easily stay up-to-date with upstream changes.
The comments suggest that getting the models to work "out-of-the-box" without any customization is unlikely. One community member notes that at a minimum, a query wrapper prompt will need to be set up to format the prompt properly for the model. Another community member indicates they have tried this with the Zephyr model, but it still struggles.
There is no explicitly marked answer in the provided information.