A typed-classification response — Layer A serializable data.
iex> a = ALLM.ClassificationAnswer.new(type: :choice, choice: "billing", confidence: 0.88)
iex> resp = ALLM.ClassificationResponse.new(answers: %{"department" => a})
iex> ALLM.ClassificationResponse.answer(resp, :department).choice
"billing"Answers
:answers is keyed by the same question ids the request used, one
ALLM.ClassificationAnswer per question. Use answer/2 to read one.
Usage
:usage is an ALLM.Usage, never nil. An adapter populates
:input_tokens and :output_tokens from the provider and sets
:total_tokens to their sum. The cost fields stay nil: no model
catalog carries classification pricing, but pricing is per input token,
so a caller who knows the rate can compute the cost.
Identifiers
:id is the provider's own identifier for the call — the value to quote
to the provider's support team — or nil when the provider sends none.
:request_id is ALLM's own correlation id. The two are never mixed: the
provider's id is not copied into :request_id, and it is not put in
:metadata, which is caller-owned.
Other fields
:model is the model the provider reports as having answered. For a
provider with moving aliases this is the pinned version, not the alias
that was sent. :provider is the provider atom. :raw carries the same
caller-responsibility contract as ALLM.Response.raw: a
non-JSON-encodable :raw raises at encode time.
Summary
Functions
The answer for id, or nil. Atom ids are stringified.
Build a %ClassificationResponse{} from keyword opts.
Types
@type t() :: %ALLM.ClassificationResponse{ answers: %{required(String.t()) => ALLM.ClassificationAnswer.t()}, id: String.t() | nil, metadata: map(), model: String.t() | nil, provider: atom() | nil, raw: term(), request_id: String.t() | nil, usage: ALLM.Usage.t() }
Functions
@spec answer(t(), String.t() | atom()) :: ALLM.ClassificationAnswer.t() | nil
The answer for id, or nil. Atom ids are stringified.
Examples
iex> a = ALLM.ClassificationAnswer.new(type: :yes_no, yes_probability: 0.93)
iex> resp = ALLM.ClassificationResponse.new(answers: %{"refund" => a})
iex> ALLM.ClassificationResponse.answer(resp, "refund").yes_probability
0.93
iex> ALLM.ClassificationResponse.answer(resp, "missing")
nilBuild a %ClassificationResponse{} from keyword opts.
Unknown keys raise KeyError via struct!/2.
Examples
iex> resp = ALLM.ClassificationResponse.new(model: "jev-1.13.0")
iex> {resp.answers, resp.usage}
{%{}, %ALLM.Usage{}}