Motivation :
The goal of the project was to adapt large language models for the Arabic language and create a new state-of-the-art Arabic LLM. Due to the scarcity of Arabic instruction fine-tuning data, not many LLMs have been trained specifically in Arabic, which is surprising given the large number of Arabic speakers.
Our final model was trained on a high-quality instruction fine-tuning (IFT) dataset, generated synthetically and then evaluated using the Hugging Face Arabic leaderboard.
Training :
This model is the 2B version. It was trained for 2 days on 1 A100 GPU using LoRA with a rank of 128, a learning rate of 1e-4, and a cosine learning rate schedule.
Evaluation :
| Metric | Slim205/Barka-2b-it |
|---|---|
| Average | 46.98 |
| ACVA | 39.5 |
| AlGhafa | 46.5 |
| MMLU | 37.06 |
| EXAMS | 38.73 |
| ARC Challenge | 35.78 |
| ARC Easy | 36.97 |
| BOOLQ | 73.77 |
| COPA | 50 |
| HELLAWSWAG | 28.98 |
| OPENBOOK QA | 43.84 |
| PIQA | 56.36 |
| RACE | 36.19 |
| SCIQ | 55.78 |
| TOXIGEN | 78.29 |
Please refer to https://github.com/Slim205/Arabicllm/ for more details.
- Downloads last month
- 780