Bassel Koshak

Artificial Intelligence Specialization · Cohort 2024

Augmenting Phi-3 for Arabic: Granting LLMs Multilingual Capability with Minimal Training Time

This project introduces a novel method for expanding the vocabulary of pre-trained Large Language Models (LLMs) for Arabic without the need for costly full pre-training. By integrating vocabulary expansion, Low-Rank Adaptation (LoRA), and targeted fine-tuning, we aim to boost model performance while minimizing computational resources. This approach sets a new benchmark for efficient language model enhancement in low-resource settings.

Read the final report (PDF)

See it in the program report (PDF)