Classify Text with a Loaded Local Model

Run a local text-classification pipeline, requesting the top label for each text. Prediction never downloads a model or falls back to hosted inference. Input order and duplicates are preserved. Missing and empty-string texts retain missing labels and scores without being sent to Python. Zero-length and entirely missing or empty inputs do not call the backend.

Usage

hf_classify_local(text, model, batch_size = 32L, truncation = TRUE)

Arguments

text

Character vector of texts to classify.

model

An hf_local_model handle loaded with task = "classify".

batch_size

Positive scalar integer. Python inference batch size.

truncation

Logical. Truncate long inputs to the tokenizer's maximum length. With FALSE, overlong inputs may produce a backend error.

Value

A tibble with text, character label, and numeric score, matching the schema of hf_classify().

Details

Scores are model outputs, not calibrated measures of certainty. By default, long texts are truncated to the tokenizer's supported maximum length.

See also

hf_load_local_model(), hf_embed_local()

Examples

if (FALSE) { # \dontrun{
model <- hf_load_local_model(task = "classify")
hf_classify_local(c("I like this.", NA, "I dislike this."), model)
} # }