AshArcadic.SparseVectorIndex (AshArcadic v0.1.0)

Sparse (learned-sparse / BM25-style) vector-index configuration declared in an arcade do … end block. The %AshArcadic.SparseVectorIndex{} struct is the target of the sparse_vector_index DSL entity; it is declaration only — metadata the read path reads at query time (the Type[tokens,weights] reference), plus compile-time validation (ValidateSparseVectorIndex). AshArcadic does NOT create the index — the host app runs Arcadic.Vector.create_sparse_index/5 (there is no migration/DDL machinery here).

Unlike a dense vector_index — whose name IS the vector attribute — a sparse index needs a logical name plus a (tokens, weights) attribute PAIR: tokens (an integer-array attribute of token ids) and weights (a float-array attribute of the matching weights). No dimensions/ modifier: those are host-side index-creation metadata with no query-time consumer (sparse token/ weight vectors are variable-length — there is no query-vector length to validate).

Summary

Types

t()

Types

t()

@type t() :: %AshArcadic.SparseVectorIndex{
  __spark_metadata__: term(),
  name: atom(),
  tokens: atom(),
  weights: atom()
}