Pull requests / #1671

#1671 monitor: show prefill rate in recent requests

open · @hash13 · 0 comments · View on GitHub

BenchmarksServer & API

Description

## Summary
Add prefill throughput to the **Monitor → Recent requests** table, making it easier to compare prompt processing and decode performance for completed requests.

## What changed
- `serve/web/index.html`
  - Added a **Prefill t/s** column.
  - Renamed **Tok/s** to **Decode t/s** for clarity.
  - Added a tooltip explaining the prefill metric.
- `serve/web/app.js`
  - Calculate prefill throughput using newly processed prompt tokens and engine-reported prompt time.
  - Exclude reused cache tokens from the calculation.
  - Round the displayed rate to whole tokens per second, consistent with the existing monitor tile.
  - Display `–` when valid timing data is unavailable.
- `serve/web/app.css`
  - Prevent line wrapping in the recent requests table, including column headers and status badges.
  - Reduce horizontal cell padding to 8px to accommodate the additional column.
  - Preserve horizontal scrolling as a fallback on narrower screens.

## Extra Notes
- Frontend-only change; no backend or engine modifications.
- Uses metrics already available in completed request history.

**Screenshot**

<img width="1211" height="1004" alt="grafik" src="https://github.com/user-attachments/assets/c862cb0b-fd06-485f-937a-f0832bfd0b97" />

Related on strata.com

Editorial links to help you install, pick models, or read release notes — not part of the upstream thread.