Skip to content

Web Studio & Streamlit App (infersize ui)

🚀 Live Cloud Instance: You can test the Web Studio directly in your browser without local installation at infersize.streamlit.app.

For interactive capacity planning, architecture design meetings, and visual RFC reviews, infersize includes a standalone Streamlit Web Studio.

1. Launching the Studio

Ensure the UI extra is installed: 

pip install "infersize[ui]"

Launch the interactive studio: 

# Direct CLI shortcut
infersize ui

# Or via standard streamlit
streamlit run app.py

The browser will open automatically at http://localhost:8501.

2. Interactive Features

  • Real-Time Sliders:
  • Adjust sequence length (\(1\text{K}\text{--}128\text{K}\) tokens).
  • Adjust concurrency batch size (\(1\text{--}64\) active streams).
  • Toggle Tensor Parallelism (\(TP = 1, 2, 4, 8\)).
  • Change weight precision (FP16, INT8, INT4, FP8, BitNet).
  • Multi-GPU Comparison Matrix:
  • Live comparison table across RTX 3050, RTX 4090, A10G, L40S, A100 (40GB/80GB), and H100.
  • Automatic badge indicators: [FIT], [OOM], and [RIGHT-SIZED].
  • Architecture RFC Exporter:
  • One-click download of a production-ready Markdown report (architecture_rfc.md) detailing memory bounds, roofline decode throughput, and cloud TCO estimates ($ / 1M tokens) for engineering stakeholders.