MIPROv2 Teleprompter — MIPROv2 • dsprrr

MIPROv2 jointly optimizes instructions and few-shot demonstrations using a discrete Bayesian optimization loop with minibatch evaluation for a root Predict module. For graphs with nested predictors such as RLM, it optimizes child instructions only and requires max_bootstrapped_demos = 0L; nested demo bootstrapping fails explicitly until predictor-local evidence is available.

Usage

MIPROv2(
  metric = NULL,
  metric_threshold = NULL,
  max_errors = 5L,
  task_model = NULL,
  teacher_settings = NULL,
  max_bootstrapped_demos = 4L,
  max_labeled_demos = 4L,
  auto = "light",
  num_candidates = NULL,
  num_threads = 1L,
  seed = 9L,
  track_stats = TRUE,
  log_dir = NULL
)

Arguments

metric: A metric function for evaluating predictions (required).
metric_threshold: Minimum score required for acceptance.
max_errors: Maximum number of errors allowed during optimization.
task_model: Optional ellmer Chat used to evaluate tasks. NULL uses the .llm supplied to compile().
teacher_settings: List of settings for the teacher model.
max_bootstrapped_demos: Maximum number of bootstrapped demonstrations.
max_labeled_demos: Maximum number of labeled demonstrations.
auto: Auto-tuned settings: "light", "medium", "heavy", or NULL.
num_candidates: Optional override for number of instruction candidates.
num_threads: Number of threads to use for evaluation.
seed: Random seed for reproducibility.
track_stats: Whether to track trial history.
log_dir: Directory for trial logging.

Value

A MIPROv2 teleprompter object.

Examples


if (FALSE) { # \dontrun{
tp <- MIPROv2(
  metric = metric_exact_match(field = "answer"),
  auto = "light",
  max_bootstrapped_demos = 4L
)

qa_module <- module(signature("question -> answer"))
trainset <- data.frame(question = "Capital of France?", answer = "Paris")
valset <- data.frame(question = "Capital of Japan?", answer = "Tokyo")
llm <- ellmer::chat_openai()
compiled <- compile(qa_module, tp, trainset, valset = valset, .llm = llm)
} # }