inceptionai/Jais-2-8B-Chat

Jais-2: The Next Generation of Arabic Frontier LLMs

Model Overview

Jais-2-8B-Chat is a bilingual Arabic–English language model developed by MBZUAI, Inception, and Cerebras. Jais-2-8B-Chat Model is trained from scratch on Arabic and English data and is powered by a custom Arabic-centric vocabulary, it efficiently captures Modern Standard Arabic, regional dialects, and mixed Arabic–English code-switching. The model is openly available under a Apache 2.0 license and also deployed as a fast, production-ready chat experience running on Cerebras hardware. Visit the Jais-2 Web App.

Key Technical Specifications

• Model Developers: MBZUAI, Inception, Cerebras.
• Languages: Arabic (MSA & dialects) and English
• Architecture: Transformer-based, Decoder-only architecture with multi-head self-attention.
• Parameters: 8 Billion
• Context Length: 8,192
• Vocabulary Size: 150,272
• Training Infrastructure: Optimized for Cerebras CS-2 and Condor Galaxy clusters
• Key Design Choices: Rotary Position Embeddings (RoPE), Squared-ReLU activation, custom μP parameterization, and 8:1 filter-to-hidden size ratio.

---

How to Use the Model

Using Transformers

1. Clone the Jais-2 compatible Transformers fork

# Pull the latest version and ensures you have the most up-to-date features/models and bug fixes.
# Note: could be not as stable as an official PyPI release.
uv pip install git+https://github.com/huggingface/transformers.git 

2. Load and Inference on the Model

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model and tokenizer
model_name = "inceptionai/Jais-2-8B-Chat"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")

# Example Arabic prompt
system_prompt = "أجب باللغة العربية بطريقة رسمية وواضحة."
user_input = "ما هي عاصمة الإمارات؟"

# Apply chat template (always)
chat_text = tokenizer.apply_chat_template(
    [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_input}
    ],
    tokenize=False,
    add_generation_prompt=True
)

# Tokenize and generate
inputs = tokenizer(chat_text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=8192, temperature=0)

# Decode and print
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
#عاصمة الإمارات العربية المتحدة هي أبوظبي.

Using vLLM

1. Clone the Jais 2–compatible vLLM fork

# Pending PR merge to the official package
git clone --branch jais2 --single-branch \
    https://github.com/inceptionai-abudhabi/vllm.git
cd vllm
uv pip install -e . # If you install vllm after transformers, please re-install transformers again from this branch: https://github.com/inceptionai-abudhabi/transformers.git

2. Load and Inference on the Model

from vllm import LLM, SamplingParams

# Load model and tokenizer
model_name = "inceptionai/Jais-2-8B-Chat"
llm = LLM(model=model_name, tensor_parallel_size=1)
tokenizer = llm.get_tokenizer()

# Example Arabic prompt
system_prompt = "أجب باللغة العربية بطريقة رسمية وواضحة."
user_input = "ما هي عاصمة الإمارات؟"

# Apply chat template (always)
chat_text = tokenizer.apply_chat_template(
    [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_input}
    ],
    tokenize=False,
    add_generation_prompt=True
)

# Run generation
sampling_params = SamplingParams(max_tokens=8192, temperature=0)
outputs = llm.generate([chat_text], sampling_params)

#Print output
print(outputs[0].outputs[0].text)
#عاصمة الإمارات العربية المتحدة هي أبوظبي.

Or serve through command line (CLI)

vllm serve inceptionai/Jais-2-8B-Chat \
    --served-model-name inceptionai/Jais-2-8B-Chat-Local --dtype bfloat16 \
    --tensor-parallel-size 1 --max-model-len 8192 --max-num-seqs 256 \
    --host 0.0.0.0 --port 8042 --api-key "Optional"

---

Evaluation

Performance Overview

We evaluate Jais-2-8B across two key benchmarks that capture both instruction following and generative Arabic ability: IFEval (English and Arabic) and AraGen-12-24 (3C3H).

IFEval Results (Strict 0-shot)

Model Name	En-Prompt	En-Instruction	Ar-Prompt	Ar-Instruction
Qwen2.5-7B-Instruct	54.31	71.65	46.04	55.85
Qwen3-8B	74.90	80.72	58.66	67.09
gemma-2-9b-it	66.27	75.73	48.51	58.07
Llama-3.1-8B-Instruct	67.06	77.01	39.85	47.63
aya-expanse-8b	54.31	65.39	45.54	56.49
c4ai-command-r7b-12-2024	68.24	76.88	52.72	61.39
c4ai-command-r7b-arabic-02-2025	75.88	80.84	62.38	70.57
ALLaM-7B-Instruct-preview-v1	51.76	62.45	45.54	53.80
ALLaM-7B-Instruct-preview-v2	56.90	66.20	39.10	46.20
Fanar-1-9B-Instruct	55.69	65.26	48.27	58.39
Falcon-H1-7B-Instruct	77.06	83.397	31.93	35.44
jais-family-6p7b-chat	26.70	37.70	22.50	32.10
jais-adapted-7b-chat	36.90	49.30	22.50	33.90
Jais-2-8B (ours)	63.14	72.80	58.17	67.09

---

AraGen-12-24 (3C3H) Results

Model Name	3C3H Score (%)	Correctness	Completeness	Conciseness	Helpfulness	Honesty	Harmlessness
Fanar-1-9B-Instruct	53.16	61.53	60.90	18.14	57.71	59.15	61.53
ALLaM-7B-Instruct-preview-v1	53.16	61.41	58.30	23.27	55.73	58.93	61.32
ALLaM-7B-Instruct-preview-v2	51.86	63.24	59.06	15.27	53.07	57.67	52.86
gemma-2-9b-it	51.74	58.90	58.90	18.34	57.97	57.44	58.90
c4ai-command-r7b-arabic-02-2025	49.18	56.83	56.47	14.36	54.74	56.00	56.65
aya-expanse-8b	48.29	56.12	56.12	11.72	54.68	55.19	55.94
Qwen2.5-7B-Instruct	47.46	54.60	54.48	15.59	52.33	53.20	54.57
Falcon-H1-7B-Instruct	47.28	56.44	55.81	18.34	44.73	52.59	55.78
c4ai-command-r7b-12-2024	44.05	51.44	50.96	13.04	48.29	49.22	51.35
jais-family-6p7b-chat	41.00	47.55	47.31	12.43	45.22	45.97	47.55
jais-adapted-7b-chat	39.42	46.36	44.09	15.32	40.62	43.79	46.36
Llama-3.1-8B-Instruct	37.83	44.21	44.09	14.16	39.67	40.65	44.21
Qwen3-8B	36.52	43.49	42.77	7.14	41.43	41.19	43.13
Jais-2-8B (ours)	58.64	68.94	68.10	11.83	66.88	67.20	68.88

Overall, our results show that:

• Jais-2-8B delivers competitive Arabic and English instruction-following performance across IFEval.
• Jais-2-8B achieves the highest scores across nearly all AraGen metrics, outperforming Fanar-1-9B-Instruct and ALLaM-7B on Arabic generative tasks.

---

Intended Use

Target Audiences

• Academics: Researchers focusing on Arabic NLP, multilingual modeling, or cultural alignment
• Businesses: Companies targeting Arabic-speaking markets
• Developers and ML Engineers: Integrating Arabic language capabilities into applications and workflows

Appropriate Use Cases

• Research:

  • Natural language understanding and generation tasks
  • Conducting interpretability or cross-lingual alignment analyses
  • Investigating Arabic linguistic or cultural patterns

• Commercial Use:

  • Building chat assistants for Arabic-speaking audiences
  • Performing sentiment and market analysis in regional contexts
  • Summarizing or processing bilingual Arabic–English documents
  • Creating culturally resonant Arabic marketing and entertainment content for regional audiences

Inappropriate Use Cases

• Harmful or Malicious Use:

  • Producing hate speech, extremist content, or discriminatory language
  • Creating or spreading misinformation or deceptive content
  • Engaging in or promoting illegal activities

• Sensitive Information:

  • Handling or generating personal, confidential, or sensitive information
  • Attempting to infer, reconstruct, or guess sensitive information about individuals or organizations

• Language Limitations:

  • Applications requiring strong performance outside Arabic or English languages

• High-Stakes Decisions:

  • Making medical, legal, financial, or safety-critical decisions without human oversight

Citation

If you find our work helpful, please give us a cite.

@techreport{jais2_2025,
  title        = {Jais 2: {A} Family of {A}rabic-Centric Open Large Language Models},
  author       = {
        Anwar, Mohamed and 
    Freihat, Abdelhakim and 
    Ibrahim, George and 
    Awad, Mostafa and 
    Sadallah, Abdelrahman Atef Mohamed Ali and 
    Gosal, Gurpreet and 
    Ramakrishnan, Gokul and 
    Mishra, Biswajit and 
    Chandran, Sarath and 
    Frikha, Ahmed and 
    Joshi, Rituraj and 
    Goffinet, Etienne and 
    Maiti, Abhishek and 
    El Filali, Ali and 
    Al Barri, Sarah and 
    Ghosh, Samujjwal and 
    Pal, Rahul and 
    Mullah, Parvez and 
    Shukla, Awantika and 
    Siddiki, Sajid and 
    Kamboj, Samta and 
    Pandit, Onkar and 
    Sahu, Sunil and 
    El Badawy, Abelrahman and 
    Mohamed, Amr and 
    Chamma, Ahmad and 
    Dufraisse, Evan and 
    Bounhar, Abdelaziz and 
    Bouch, Dani and 
    Abdine, Hadi and 
    Shang, Guokan and 
    Koto, Fajri and 
    Wang, Yuxia and 
    Xie, Zhuohan and 
    Mekky, Ali and 
    Elbadry, Rania Hossam Elmohamady and 
    Ahmad, Sarfraz and 
    Ahsan, Momina and 
    El-Herraoui, Omar Emad Mohamed and 
    Orel, Daniil and 
    Iqbal, Hasan and 
    Elzeky, Kareem Mohamed Naguib Abdelmohsen Fahmy and 
    Abassy, Mervat and 
    Ali, Kareem and 
    Eletter, Saadeldine and 
    Atif, Farah and 
    Mukhituly, Nurdaulet and 
    Li, Haonan and 
    Han, Xudong and 
    Singh, Aaryamonvikram and 
    Quraishi, Zain and 
    Sengupta, Neha and 
    Murray, Larry and 
    Sheinin, Avraham and 
    Hestness, Joel and 
    Vassilieva, Natalia and 
    Ren, Hector and 
    Liu, Zhengzhong and 
    Vazirgiannis, Michalis and 
    Nakov, Preslav
  },
  institution  = {IFM},
  type         = {Technical Report},
  year         = {2025},
  month        = dec,
  day          = {09},
}