seismic

Module for indexing and searching using Seismic.

This module provides the classes and configurations for indexing and searching using Seismic in the Lightning IR framework.

Seismic accelerates search for learned sparse models by restructuring how the traditional inverted index stores and processes data. Instead of evaluating every single document individually, it divides the inverted lists of terms into manageable blocks and precomputes maximum score upper bounds for each of these blocks. During a search, the system compares the query against these precalculated bounds to quickly evaluate entire groups of documents at once. If a block’s maximum possible score falls below the threshold needed to make it into the top results, Seismic safely skips that entire block without calculating any individual document scores.

Seismic: Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations

Modules

seismic_format	SeismicFormatConverter class for converting embeddings to a seismic format.
seismic_indexer	SeismicIndexer class for indexing documents using the Seismic library.
seismic_searcher	Seismic Searcher for Lightning IR Framework