1 Vectors
(vector-dot a b) -> flonum? |
(vector-norm v) -> flonum? |
(vector-cosine a b) -> flonum? |
(vector-normalize v) -> flvector? |
a, b, v : flvector? |
Dimension mismatches and zero vectors raise: a zero embedding means the embedder produced garbage upstream, and silently scoring it would corrupt every ranking. Normalization is idempotent. Entries are normalized once at add time, so search costs one dot product per entry — no norms are recomputed per query.