Skip to content

About

Complete Pinokio package for running SmolLM3-3B locally with Gradio interface. Features GPU acceleration, extended thinking mode, and complete privacy.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

27 Commits

Folders and files

Repository files navigation

SmolLM3-3B Chatbot (Pinokio)

Local chat UI for SmolLM3-3B using a Gradio web app. The model and tokenizer load on your machine; nothing is sent to external APIs beyond Hugging Face Hub downloads for the model weights (unless you change that in code).

What it does

  • Installs Python dependencies and PyTorch (via torch.js for your OS/GPU).
  • Starts a Gradio server that prints a local URL; Pinokio captures that URL for an Open Web UI action.
  • Supports optional Save Disk Space (link.js) to deduplicate the virtual environment.

Requirements

  • Pinokio with the ai bundle (installed automatically by install.js).
  • transformers 4.53.0 or newer — the SmolLM3 architecture is not available in older releases.
  • Roughly 6 GB of free disk space for the model weights, downloaded from the Hugging Face Hub on first run.

How to use (Pinokio)

  1. Open this project in Pinokio.
  2. Run Install once.
  3. Run Start and use Open Web UI when it appears, or open the Terminal tab for logs.

Update runs git pull and reinstalls dependencies. Reset removes the virtual environment(s) so you can reinstall cleanly; it does not delete application source under app/.

Programmatic access

JavaScript (Pinokio scripts)

Launcher scripts live in the project root: install.js, start.js, update.js, reset.js, link.js, torch.js. The UI is defined in pinokio.js. From another Pinokio script you can start flows with script.start (see Pinokio PINOKIO.md).

Python

Run the app directly after activating the same venv Pinokio uses (env inside app/):

# Windows (example)
cd app
env\Scripts\activate
python app.py --host 127.0.0.1 --port 7860

Available flags: --host, --port, and --share (creates a public Gradio link). The model is loaded only after the arguments are parsed, so python app.py --help returns immediately without downloading weights.

cURL

With the server running at http://127.0.0.1:<port>, use Gradio’s HTTP API. Fetch the OpenAPI schema from the running app (Gradio exposes API docs) or use gradio_client from Python for typed calls.

Example (replace PORT with the port shown in the terminal):

curl -s "http://127.0.0.1:PORT/" -o /dev/null

For generation, prefer the Gradio Client library or the interactive UI; exact JSON routes depend on your Gradio version.

Layout

  • app/ — application code (app.py, requirements.txt) and the env/ virtual environment used to run it.
  • Root — Pinokio launcher scripts (install.js, start.js, update.js, reset.js, link.js, torch.js, pinokio.js).

About

Complete Pinokio package for running SmolLM3-3B locally with Gradio interface. Features GPU acceleration, extended thinking mode, and complete privacy.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages