Generation Stats

A row of small metric pills for an assistant turn — output token count, throughput (tokens/second) and latency. It's purely presentational and data-driven: pass the numbers you have and it renders only those, in a compact bg-muted pill row that matches the rest of the chat chrome. When there's no data at all, it renders nothing.

Throughput is derived from tokens + latencyMs when you don't pass tokensPerSecond explicitly. Unit labels are configurable (labels) so the component carries no copy of its own — the defaults are English (tokens / tok/s / ms).

Import

1import { GenerationStats } from "@yuhuanowo/yunui/chat";

Example

The first row passes tokens + latencyMs (throughput is derived); the second overrides the labels for a localized look and passes an explicit tokensPerSecond.

Usage

1<GenerationStats
2  tokens={message.tokens?.output}
3  latencyMs={message.latencyMs}
4  labels={{ tokens: t("tokens"), speed: t("tokensPerSecond"), latency: t("ms") }}
5/>

Pair it with the assistant name/model chip in a ChatMessage badges slot, and with ThinkingBlock (isStreaming renders the "thinking…" state) for the reasoning indicator.

Props