Creates a parsnip model specification for LLM-based prediction using dsprrr.
Usage
llm_predict(
mode = "classification",
signature = NULL,
temperature = NULL,
top_p = NULL,
model = NULL,
provider = NULL
)Arguments
- mode
Model mode, typically "classification" or "regression" (for text tasks, classification is most common).
- signature
A dsprrr signature string or Signature object.
- temperature
Temperature parameter for LLM (tune-able).
- top_p
Top-p parameter for LLM (tune-able).
- model
LLM model name (e.g., "gpt-4o-mini").
- provider
LLM provider (e.g., "openai", "anthropic").
Examples
if (FALSE) { # \dontrun{
library(parsnip)
library(tune)
# Create LLM model spec
llm_spec <- llm_predict(
mode = "classification",
signature = "text -> sentiment: enum('positive', 'negative', 'neutral')"
) |>
set_engine("dsprrr", model = "gpt-4o-mini")
# With tunable parameters
llm_spec_tuned <- llm_predict(
mode = "classification",
signature = "text -> sentiment",
temperature = tune()
) |>
set_engine("dsprrr")
} # }