A compact, efficiency-focused model built for on-device deployment. It runs on mobile hardware through quantization-aware training, which keeps the footprint small while preserving reasonable text generation quality. The trade-off is clear: you get portability and local execution, not the depth of larger server-side models.