Tune.Prepared (Tune v0.5.0)

Copy Markdown View Source

A rendered, inspectable request for one predictor call.

It holds normalized inputs, the signature, the demonstrations used, the adapter and its options, the ReqLLM context, and the output descriptor. It holds no model, credentials, or request options, and preparing it makes no model request. A host may execute context with output through its own inference path and pass the response to Tune.Predictor.finish/2.

Summary

Types

t()

@type t() :: %Tune.Prepared{
  adapter: module(),
  adapter_options: keyword(),
  context: ReqLLM.Context.t(),
  demos: [Tune.Demo.t()],
  inputs: map(),
  output: ReqLLM.Output.t(),
  signature: Tune.Signature.t()
}