PromptEvalCall
One evaluation Call of a run (1.63.0): which scenario, arm, repetition and family it ran, how it ended, its Turn metrics, and its transcript under include=transcript. A calibration's labeller reads the transcript here.
An object with 14 fields.
Fields
| Field | Type | Required | Description |
|---|---|---|---|
arm | PromptEvalArm | yes | |
call_id | CallId | yes | |
ended_at | Timestamp or null | yes | |
ended_reason | EndedReason or null | yes | |
key | PromptEvalKey or null | yes | The key of an inline test; else null. |
model_family | ModelFamily or null | yes | The family the Call ran under; null for the family of the chain's primary. |
object | "prompt_eval_call" | yes | |
repetition | integer | yes | |
run_id | PromptEvalRunId | yes | |
scenario_id | PromptEvalScenarioId or null | yes | Null for an inline test. |
started_at | Timestamp or null | yes | |
suite_id | PromptEvalSuiteId | yes | |
transcript | array or null | yes | The text of each Turn, the content class transcript: null unless the read names include=transcript, and null once scrubbed or for an older worker. |
turns | array of TurnMetrics | yes | The Turn metrics, without text, Turn 0 first; empty for an older worker. |
Used by
The operations whose parameters, request body or responses name this shape directly.