Mini Indig LLM Starter Kit for Your Own Language Revival Project

Énska Floral 3D Asset by MoniGarr for MohawkLanguage.ca

Let’s Bring Our Languages Back to Life

For over 30 years now, I’ve been producing Kanien’kéha (Mohawk Language) revival and retention tools — interactive media, AI chatbots, learning games, mobile apps, aiml knowledge bases, workshops, voice tools, and now LLMs.

Here we are today speaking, singing, coding, remembering, and regenerating: The Mini Indig LLM Starter Kit.

monigarr/mini-indig-llm-kit: mini-indig-llm-kit : Polysynthetic Language Mini-LLM Starter Kit is an open-source project that fine-tunes LLaMA 3 (8B) with QLoRA and a custom tokenizer, tailored for morphologically complex languages

What This Project Does (and Why It Matters)

The mini-indig-llm-kit is an open-source, community-first AI toolkit designed to help you build your own language model for your Indigenous language — especially if your language:

• Doesn’t have an official ISO code for each dialect
• Doesn’t have much written data online
• Is morphologically complex (like most polysynthetic languages)
• Is alive and deserves to be spoken today & tomorrow instead of forgotten, preserved, quietly fading on paper and gathering dust on a shelf.

This project fine-tunes Meta’s LLaMA 3 (8B) model using QLoRA for efficiency. You can run it offline, even on older laptops or low-spec machines. The kit is designed to be accessible, respectful, and powerful — so language teams of all kinds can build real, working language models together.

Features That Make It Powerful

This isn’t just a research prototype. It’s a living system you can grow with your community:

• Jupyter Notebooks for training, inference, evaluation, visualization
• Editable templates for your corpus, tokenizer rules, ethical consent files
• Low-spec compatible code — CPU fallback, offline-first deployment
• Worksheets for non-coders (artists, Elders, teachers, youth)
• Ethical frameworks for consent, data sovereignty, and cultural respect
• Public GitHub repo with clean structure and contribution guide

Whether you’re an Indigenous linguist, a language learner, a teacher, a speaker or a developer — this toolkit is for you.

Preservation vs Revival: Let’s Be Clear

I want to make this distinction because it matters:

🔸 Preservation is when we archive our languages — grammars, dictionaries, recordings — to prove they existed. It’s about saving what’s left. My community is fortunate in that we have / had many fluent speakers who produced a wide variety of Kanien’kéha content (print, audio, video). I recently digitized Kanien’kéha recordings from the 1940s that were found on the old RCA Victor giant tape cassettes. 🔸 Revival & Retention is about using our languages in daily life — in the kitchen, on TikTok, in the classroom, in boardrooms, in VR, in video games, in ceremony, on YouTube, on Television, on the radio, on our community stop signs and traffic signs.

This project supports revival and retention — making your language usable, vibrant, and integrated into modern digital life while supporting and promoting respect for cultural protocols, fully informed prior consent with the community fluent speakers / elders / people.

How to Use the Mini Indigenous LLM Starter Kit

Here’s how you can get started:

1. Clone the Repo

git clone https://github.com/monigarr/mini-indig-llm-kit.git
cd mini-indig-llm-kit

2. Install the Dependencies

The README explains how to set up a virtual environment and install required packages. This kit is designed to run on CPU or GPU — even on air-gapped laptops.

pip install -r requirements.txt

3. Train or Fine-tune

Use the Jupyter notebooks inside notebooks/ to load your own corpus, tokenizer rules, and ethical metadata. If you don’t have data yet, you can use the provided worksheets to collaborate with Elders or teachers to build it.

jupyter lab

Open:

• 1_prepare_data.ipynb
• 2_train_with_qlora.ipynb
• 3_infer_offline.ipynb

4. Evaluate & Share

Use 5_evaluation_metrics.ipynb to measure segmentation accuracy, then use the built-in GitHub Action to auto-publish your results to a team dashboard — Elders can leave corrections and feedback right in your cloned GitHub repository.

A Few Things You Can Do with Your Language Model

• Build a voice assistant for home or classroom use (offline)
• Translate stories and oral history for community archives
• Create auto-correct / grammar suggestion tools for youth
• Power interactive art, games, or cultural installations
• Enable simple chatbot apps to help learners explore your language
• Explore language patterns and preserve rare dialects through AI simulation

Want Help Building Yours?

I work as a technical consultant and AI systems architect specializing in:

• Indigenous language AI
• Polysynthetic morphologies
• Ethical data governance
• Offline-first design
• Community-led research

If your community or organization is building a language revival project and wants expert help — I’m available. I love building custom versions of this kit and leading workshops to get teams up and running.

Reach out: monigarr@monigarr.com Hire me: MoniGarr.com