---
title: LFM2.5-VL-1.6B WebGPU
emoji: 🧠
colorFrom: purple
colorTo: blue
sdk: docker
pinned: false
models:
  - LiquidAI/LFM2.5-VL-1.6B
  - LiquidAI/LFM2.5-VL-1.6B-ONNX
short_description: In-browser vision-language inference with LFM2.5-VL-1.6B
---

# LFM2.5-VL-1.6B WebGPU Demo

In-browser vision-language inference with LFM2.5-VL-1.6B, powered by ONNX Runtime and WebGPU.

Everything runs entirely in your browser with WebGPU acceleration - no data is sent to a server.

## Features

- **Live Webcam Captioning**: Stream from your webcam with real-time AI-generated captions
- **Multiple Quantization Options**: Choose between Q4/Q4 (~1.5 GB), Q4/FP16 (~2.3 GB), or FP16/FP16 (~3.2 GB)
- **Browser Caching**: Models are cached locally after first download for faster subsequent loads
- **Adjustable Resolution**: Configure capture resolution (256-512px) for performance tuning

## Requirements

- WebGPU-enabled browser
- ~2-4 GB memory depending on quantization choice
- Webcam access for live captioning