responses

Per-call response metadata, one entry per underlying model call (n-batching).

Since

0.3.0-beta01