Client-side demo tool. Model runs in browser using transformers.js. First load can take time.

Initializing...
If WebGPU fails, fallback to WASM.
Model state: initializing

Any compatible Hugging Face model can be used. Recommendation adapts to your device capability.

Tip: Browser inference usually needs ONNX-ready repos (for example onnx-community/*).

If an experimental model fails, the tool automatically falls back to a stable Qwen2.5 model.

Downloaded model files are cached in your browser for faster future loads on this device.