InferFlux — LLM Inference System Visualizations

English · 简体中文 · Español

A collection of interactive, single-file web animations for LLM inference systems. Each page is standalone and dependency-free — drag the parameters to watch system behavior change in real time.

Online · Live: https://Abatom.github.io/inferflux/

Pages

• index.html — homepage, entry point to all sections
• pd-disagg.html — Prefill/Decode disaggregation simulator Faithfully reproduces SGLang's PD-disaggregated request lifecycle, with live TTFT / TPOT / throughput / queueing / KV utilization.
• calc-input.html — input throughput calculator (prefill, MFU / roofline)
• calc-output.html — output throughput calculator (decode, MBU / TPOT)

Usage

Open index.html in a browser. No install, no build, no server.

Every page is a self-contained static HTML file (CSS + DOM + JS) — copy one file and it just works.

Run locally

For local access over HTTPS (some pages require a secure context), first generate a self-signed cert (not included in the repo), then start the bundled HTTPS server:

# generate a self-signed cert
openssl req -x509 -newkey rsa:2048 -nodes -days 365 \
  -keyout server/key.pem -out server/cert.pem -subj "/CN=localhost"

# start
python3 server/https_server.py

Then open https://localhost:8888/ in your browser (self-signed cert, accept the warning on first visit).

License

Licensed under CC BY-NC 4.0 (Attribution-NonCommercial). Free to use, modify, and share with attribution and for non-commercial purposes only. For commercial use, please obtain prior permission from the author. See LICENSE.