Skip to main content Module distance Copy item path Source compute_distance Dispatch distance computation based on metric. All three metrics use
SIMD-unrolled kernels. cosine_distance Cosine distance = 1 - cosine_similarity. Range [0, 2]; 0 = identical direction. cosine_similarity Cosine similarity between two vectors.
Returns 1.0 for identical direction, -1.0 for opposite, 0.0 for orthogonal.
Returns 0.0 when either vector is zero (avoids divide-by-zero). cosine_similarity_simd Cosine similarity using 8-element loop unrolling. dot_product Dot product of two vectors. dot_product_distance Dot-product distance = 1 - dot_product(a, b).
NOTE: Only meaningful as a proper distance metric for unit-normalized vectors. dot_product_simd Dot product using 8-element loop unrolling. euclidean_distance Euclidean (L2) distance between two vectors. euclidean_distance_simd Euclidean (L2) distance using 8-element loop unrolling. normalize Return a unit-normalized copy of v.
If magnitude < 1e-10 (near-zero vector), returns v unchanged to avoid instability. to_score Convert a compute_distance value into a similarity score in which higher is
better (the ranking/hydration convention used across all index backends).