Generate Embeddings with a Loaded Local Model

Run SentenceTransformers on a reusable local handle. Prediction never downloads a model or falls back to hosted inference. Input order and duplicates are preserved. Missing texts retain NULL embeddings and NA_integer_ dimensions; zero-length and all-missing inputs do not call Python. Empty strings are valid texts. The model's own maximum sequence length controls truncation of long texts.

Usage

hf_embed_local(
  text,
  model,
  batch_size = 32L,
  normalize = FALSE,
  progress = FALSE
)

Arguments

text

Character vector of texts to embed.

model

An hf_local_model handle loaded with task = "embed".

batch_size

Positive scalar integer. Python inference batch size.

normalize

Logical. Normalize embeddings to unit length.

progress

Logical. Display the Python encoding progress bar.

Value

A tibble with text, embedding (a list of numeric vectors), and integer n_dims, matching the schema of hf_embed().

See also

hf_load_local_model(), hf_classify_local()

Examples

if (FALSE) { # \dontrun{
model <- hf_load_local_model(task = "embed")
hf_embed_local(c("Hello", NA, "Hello"), model, normalize = TRUE)
} # }