license: mit
datasets:
- Slim205/Barka_data_2B
language:
- ar
base_model:
- google/gemma-2-2b-it
Welcome to Barka-2b-it : The best 2B Arabic LLM
Motivation :
The goal of this 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 :
My model is now on the Arabic leaderboard.
| Metric | Average | ACVA | AlGhafa | MMLU | EXAMS | ARC Challenge | ARC Easy | BOOLQ | COPA | HELLAWSWAG | OPENBOOK QA | PIQA | RACE | SCIQ | TOXIGEN |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Slim205/Barka-2b-it | 46.98 | 39.5 | 46.5 | 37.06 | 38.73 | 35.78 | 36.97 | 73.77 | 50 | 28.98 | 43.84 | 56.36 | 36.19 | 55.78 | 78.29 |
Please refer to https://github.com/Slim205/Arabicllm/ for more details.
Using the Model
The model uses transformers to generate responses based on the provided inputs. Here’s an example code to use the model:
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch
model_id = "google/gemma-2-2b-it"
peft_model_id = "Slim205/Barka-2b-it"
model = AutoModelForCausalLM.from_pretrained(model_id).to("cuda")
tokenizer = AutoTokenizer.from_pretrained("Slim205/Barka-2b-it")
model1 = PeftModel.from_pretrained(model, peft_model_id)
input_text = "ما هي عاصمة تونس؟" # "What is the capital of Tunisia?"
chat = [
{ "role": "user", "content": input_text },
]
prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
outputs = model.generate(
input_ids=inputs.to(model.device),
max_new_tokens=32,
top_p=0.9,
do_sample=True
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
user
ما هي عاصمة تونس؟
model
عاصمة تونس هي تونس. يشار إليها عادة باسم مدينة تونس. المدينة لديها حوالي 2،500،000 نسمة