Labess-7b-chat is an open model instruction-tuned for Tunisian Derja, it's a continual pre-training version of jais-adapted-7b-chat with tunisian_Derja_Dataset
Below we share some code snippets on how to get quickly started with running the model. First, install the Transformers library with:
1import torch
2from transformers import pipeline
3
4pipe = pipeline(
5 "text-generation",
6 model="linagora/Labess-7b-chat-16bit",
7 model_kwargs={"torch_dtype": torch.bfloat16},
8 device="cuda" # replace with "mps" to run on a Mac device
9)
10
11messages = [
12 {"role": "user", "content": 'وين تجي تونس؟'},
13]
14
15outputs = pipe(messages, max_new_tokens=64, do_sample=True, temperature=0.2)
16assistant_response = outputs[0]["generated_text"][-1]["content"].strip()
17print(assistant_response)
1@model{linagora2025LLM-tn,
2 author = {Wajdi Ghezaiel and Jean-Pierre Lorré},
3 title = {Labess-7b-chat:Tunisian Derja LLM},
4 year = {2025},
5 month = {January},
6 url = {https://huggingface.co/datasets/linagora/Labess-7b-chat-16bit}
7}
8
Training of
Labess-7b-chat was made possible by computing AI and storage resources by GENCI at IDRIS thanks to the grant 2024-AD011014561 on the supercomputer Jean Zay’s A100 partition.
