A community member is experiencing an issue with the Faiss index in the GPT Index library, where they are encountering a TypeError related to float32 vs float64 data types. Another community member suggests that the issue may be resolved by changing the data type to float32 when adding documents and querying the index. The community members also discuss potential pull requests that may address the issue, and recommend installing the latest version of the library from the main branch to use the fix.
OK so in _add_document_to_index I changed the np.array instantiation to specify dtype=np.float32 and the problem appears to be fixed. I don't know if this will create some undesirable side effects. Things appear to work now.