@ranxianglei/pi-tps0.1.3

极简 tps 小工具：每条 assistant 回复后提示本次输出 tokens/s（minimal tokens/sec notifier for pi）

ranxianglei Published 28 days ago MIT ~666 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

No reports yet.

A tiny pi extension — 极简 tps 小工具：shows the assistant output speed after each response.

After every assistant message, pi pops a notification like:

⚡ out 342 tok · 87.3 tok/s

(out N tok = tokens generated in that message, N tok/s = average output speed over the whole message including time-to-first-token.)

# from npm
pi install npm:@ranxianglei/pi-tps

# one-off trial (no install)
pi -e npm:@ranxianglei/pi-tps

Uninstall:

pi remove npm:@ranxianglei/pi-tps

Requires a pi build where extensions are auto-discovered (default). No other dependencies.

Speed is measured per assistant message via message_start/message_end and usage.output; it is an average over the whole message, not a live per-second counter.
Rates above a sanity ceiling (default 1500 tok/s, override with PI_TPS_MAX_RATE) are physically implausible for real generation — they usually mean an instant/cached reply or a zero-length measurement window (some provider paths emit message_start milliseconds before message_end). Such values are shown as >1.5k tok/s instead of a fabricated number; non-streamed messages fall back to the message object's own timestamp.
Displayed with the theme's success color; change fg("success", ...) in extensions/tps.ts to any other token (accent, warning, ...).
If your frontend does not render footer status, the notification (ctx.ui.notify) path shown here works in both TUI and RPC/ACP mode.

MIT