Compute cosine similarity between all pairs of embeddings. Embeddings are
grouped by dimension and each group is scored with one matrix
cross-product, so the work is done by compiled linear algebra rather than
by R code for each pair. Invalid, zero-length, zero-norm, non-finite, or
dimension-mismatched pairs return `NA_real_`.
Usage
hf_similarity(embeddings, text_col = "text")
Arguments
- embeddings
A tibble with an 'embedding' column (from hf_embed).
- text_col
Character string. Name of the text column. Default: "text".
Value
A tibble with columns: text_1, text_2, similarity.
Examples
if (FALSE) { # \dontrun{
sentences <- c("I love cats", "I adore felines", "Dogs are great")
embeddings <- hf_embed(sentences)
similarities <- hf_similarity(embeddings)
} # }
