README.md

metadata

title: LFM2.5-VL-3B WebGPU
emoji: 💧
colorFrom: purple
colorTo: purple
sdk: docker
pinned: false
models:
  - LiquidAI/LFM2.5-VL-3B
  - LiquidAI/LFM2.5-VL-3B-ONNX
short_description: Run LFM2.5-VL-3B locally in your browser with WebGPU

LFM2.5-VL-3B WebGPU

Run the Q4 ONNX export of LFM2.5-VL-3B directly in a WebGPU-capable browser. Model files are downloaded from Hugging Face and cached locally; prompts, images, and generated responses remain on the user's device unless an explicitly enabled external tool is called.

Features

• Text and multiple-image conversations
• Paste, drag-and-drop, file upload, and webcam image input
• Grounding and document-layout overlays
• Built-in and MCP tool calling
• Editable system prompts and generation settings
• Persistent browser model cache
• Strict WebGPU execution with visible runtime warnings and errors

Local development

npm install
npm run dev

WebGPU requires HTTPS or localhost.

For a production-style local run:

npm run build
npm start