Decision Guide

This guide helps you navigate Lightning IR’s configuration space. It is structured as a series of decision trees: start with what you want to do, then follow the branches to pick the right model architecture, index type, loss function, and data format. Each section ends with concrete CLI and Python examples you can copy and adapt.

Page	What you will find
What Do You Want to Do?	The four top-level workflows (fit, index, search, re_rank) and a summary table of the required ingredients for each.
Which Model Architecture to Use?	Decision tree and comparison table for choosing a model architecture (DPR, SPLADE, ColBERT, MonoEncoder, SetEncoder), with quick-start code examples.
Which Index Type to Use?	Decision tree and comparison table for choosing an index and search config (FAISS variants, PLAID, Torch dense/sparse, Seismic), with quick-start code examples.
Which Loss Function to Use?	Decision tree and reference table for choosing a loss function, including knowledge-distillation and SPLADE regularization examples.
Which Dataset Format to Use?	Decision tree and reference table for the four dataset classes (TupleDataset, RunDataset, DocDataset, QueryDataset).
End-to-End Recipes	Complete end-to-end pipelines for DPR, SPLADE, and ColBERT — each covering fine-tuning, indexing, searching, and re-ranking. Includes a compatibility cheat sheet.