---
title: ZeroGPU Strip Bench
emoji: ⏱️
colorFrom: blue
colorTo: purple
sdk: gradio
python_version: "3.10"
app_file: app.py
pinned: false
---

# ZeroGPU strip-pass benchmark (throwaway test app)

Standalone benchmark to measure real timing of the vocal-isolation pass — the
same model the live Voice Remover's Step 1 uses
(`mel_band_roformer_vocals_becruily`) — on this Space's ZeroGPU hardware.

Not connected to the production site, its password gate, or its backend.
Purpose is purely to get real load-time / inference-time numbers before
deciding whether ZeroGPU is worth building into the actual product.

## Use

Upload a song, click "Run strip pass". Reports model-load time, separation
time, total, the GPU name, and an estimate of how many runs the 2-min/day free
IP-based quota buys per visitor (excludes any per-region splits on top).

## Deploy

Space hardware must be set to ZeroGPU and SDK to Gradio (already required for
`spaces.GPU` to work at all — Docker/Static Spaces can't schedule onto
ZeroGPU). Upload `app.py`, `requirements.txt`, and this `README.md` to the
Space's file root.