The-African-Research-Collective/post-training

Language Model Adapation via Direct Policy Optimization of Existing LLMs # Exploring Post Training for African Languages This repository contains code for exploring post-training across different facets for African languages. African Languages are generally low-resourced and the absense of large bodies of "high-quality" unsupervised text and instruction-response pairs means we have to be creative about how we approach the problem of training models for these languages. The main goal of this project is to explore language adaptation across different dimensions such as: - Synthesizing text in African languages: We explore popular synthetic text generation techniques and adapt them to African languages and then crank up the creativity to see what we can come up with. - Language Adaptation via alignment: We explore techniques for aligning languages and see how we can leverage these techniques to adapt models to African languages. ## Synthetic Text Generation ... ## Language Adaptation via Alignment ...