gen_ai namespace
Rius no longer defines keys in gen_ai. The deprecated keys
below are kept so a reader of older spans can tell what they were.
Generated from rius-semconv 1.5.0, the Rius attribute registry, and updated on each of its releases.
Deprecated
| Key | Type | Replaced by | Description |
|---|---|---|---|
gen_ai.first_token | event | gen_ai.response.time_to_first_chunk | The first streamed token or chunk of a generation arrived. [1] The conventions carry this signal as the span attribute gen_ai.response.time_to_first_chunk, which the SDKs emit. For a span with no first-chunk event, the Rius backend synthesizes the event it stores at ingest, at span start plus that duration and never after the span’s end; a recorded first-chunk event keeps its own time. Producers need not send it. |
Notes
[1] gen_ai.first_token: Deprecated: replaced by the gen_ai.response.time_to_first_chunk span attribute.
First streamed token/chunk arrived. The standard signal is the span
attribute gen_ai.response.time_to_first_chunk, which the SDKs emit. The
Rius backend keeps gen_ai.first_token as the name of the event it
stores. A recorded first-chunk event (OpenInference’s or the Vercel AI
SDK’s) is renamed to it and keeps its own time. For a span with none,
the backend synthesizes the event from
gen_ai.response.time_to_first_chunk, at span start plus the duration
and never after the span’s end; that clamp applies only to an event it
creates. Producers need not send it. The constant stays, deprecated,
until a major release.
Namespace exception: Predates the rule. No convention defines an event for the first streamed token.