Fast-Forward Indexes v0.8.0

Memory-mapped indexes

• OnDiskIndex now supports accessing the vectors using memory maps (setting memory_mapped=True).

Misc

• Python 3.13 is now supported.
• Aggregation (i.e., MaxP, AvgP) is now done using Pandas.
• Rankings now verify that query-document/passage pairs are unique.
• Defaults for init_size, chunk_size, and alloc_size have been updated (indexes).
• Indexes now raise an error when a vector is requested for an ID that does not exist.

API changes

• OnDiskIndex: hdf5_chunk_size has been renamed to chunk_size.
• OnDiskIndex: min_resize_val has been removed and is now inferred based on the chunk size.