ljcamargo/tachiwin_notebooks

Tachiwin notebooks for pretraining, finetuning and inference of Tachiwin models

Tachiwin: Indigenous Language Model Toolkit

Overview

Tachiwin is an open-source project developing Large Language Models (LLMs) for indigenous languages of Mexico, focusing on Tutunakú linguistic resources.

Notebooks

• Llama 3.1 8B Instruct pretraining on indigenous language corpora
• Domain-specific fine-tuning for translation and linguistic tasks
• Model deployment and inference pipeline

Google Colab Notebooks

• Pretraining Notebook
• Finetuning Notebook
• Inference Notebook

Requirements

• Python 3.10+
• PyTorch
• Transformers
• Unsloth
• Llama 3.1 8B Instruct weights

Huggingface Collection of Datasets and Models

• Tachiwin Collection

Android Application "Tachiwin" Snapshot

Fully functional app to demonstrate the translation capabilities offline or online

Features

• Multilingual support (Tutunakú/Spanish/English)
• Low-resource language model development
• Open-source linguistic technology

License

Apache 2.0

Contributors

• Luis J Camargo
• Fidencio Hernández Hernández