👀 Live Demo: LLM Poem Generator (with smolLM 135m)
info

Since this model fully run on your browser there is no server needed beyond serving the initial asset, this is private by design (You can generate any poem you wish), no data is collected, if you are in doubt just turn off your network and try the playground (without reloading).

We use a 'tiny' pre-trained Large language model named SmolLM-135M for the purpose of this demo. It was selected to allow fast download and okish speed in wasm. Remember that best performance are unlocked when running directly on hardware with GPU access (tract supports: MacOS and NVIDIA GPU). This model has been quantized in Q40 so that total size of asset is: 91Mo (model wo tokenizer being 77Mo).
mode_edit

👀 Live Demo: LLM Poem Generator

downloading model