niclasgriesshaber/Llama-3.2-3B-Instruct-bnb-4bit-lora-en-to-embg

English to Early Modern Bohemian German Translation Model

Overview

This model translates from English to Early Modern Bohemian German (EMBG). It was fine-tuned using LoRA on a unique historical dataset of 3,873 paragraph-level translation pairs sourced from legal court records. The dataset was meticulously transcribed and translated by the Chichele Professor of Economic History, Sheilagh Ogilvie, from All Souls College, University of Oxford.

Key Features

• Base Model: unsloth/Llama-3.2-3B-Instruct-bnb-4bit
• Fine-Tuning: Performed using LoRA and Unsloth, leveraging Hugging Face's Transformers and TRL libraries.
• Languages Supported:
  • Source: English
  • Target: Early Modern Bohemian German (EMBG)
• Dataset: Legal court records, manually transcribed and translated over five years. The dataset will be published in an upcoming ACL paper.

Use Cases

• Research in economic history and legal studies.
• Exploration of historical dialects and their nuances.
• Applications in language revitalisation and historical text analysis.