Mushari440/Nemotron-3-Nano-30B-sft

Nemotron-3 Nano 30B — Arabic Multitask SFT

Model Details

• Developed by: Mushari Alothman
• Model type: Causal Language Model
• Language(s): Arabic, English
• License: Apache 2.0
• Finetuned from: NVIDIA Nemotron-3 Nano 30B

This model is a Supervised Fine-Tuned (SFT) version of NVIDIA Nemotron-3 Nano 30B optimized for high-accuracy Arabic-first understanding with strong performance across multitask settings.

It is designed to produce clean, well-structured, and properly terminated responses aligned with ChatML-style training.

---

Intended Uses

Direct Use

• Arabic & English MCQ answering (A–H format)
• Context-based QA / RAG
• Instruction following
• Saudi-domain knowledge understanding

How to Use

Transformers

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "Mushari440/Nemotron-3-Nano-30B-sft"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="bfloat16",
    device_map="auto"
)

prompt = "<|im_start|>user\nما عاصمة السعودية؟\n<|im_end|>\n<|im_start|>assistant\n"

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))