Find Nearest Neighbors by Semantic Similarity — hf_nearest

Find the k most similar texts to a query text based on embedding similarity. Similarities are computed with a blocked matrix-vector product. Embeddings that are missing, non-numeric, zero-norm, non-finite, or of a different dimension than the query embedding have an `NA` similarity and sort last.

Usage

hf_nearest_neighbors(
  data,
  query,
  k = 5,
  text_col = "text",
  model = hf_default_model("embed"),
  token = NULL
)

Arguments

data: A data frame with an 'embedding' column (from hf_embed_text).
query: Character string. The query text to compare against.
k: Integer. Number of nearest neighbors to return. Default: 5.
text_col: Character string. Name of text column. Default: "text".
model: Character string. Model to use for query embedding. Should match the model used for data embeddings.
token: Character string or NULL. API token for authentication.

Value

A tibble with the k nearest neighbors, sorted by similarity (descending).

Examples

if (FALSE) { # \dontrun{
docs_embedded |>
  hf_nearest_neighbors("machine learning", k = 5)
} # }