🇦🇿 LIA-AzInstruction 🧠

✨ Fine-tuned Gemma 3 (270M) for Azerbaijani Instruction Understanding

📚 Overview

LIA-AzInstruction is a fine-tuned version of Gemma 3 (270M) trained on a custom Azerbaijani instruction dataset 🏗️
The goal is to create a compact yet intelligent model capable of understanding and responding in natural Azerbaijani language.

🧩 This project bridges the gap between global LLM architectures and local Azerbaijani AI development 🇦🇿

🧾 Dataset

🗂️ Total Samples: 400
📘 Structure: Alpaca-style (input → output)
📑 Topics covered:

#	Field	Description
🏛 1	Tarix	Ancient states, historical facts, culture
➗ 2	Riyazi Analiz	Derivatives, integrals, limits
🤖 3	Mexatronika və Robototexnika	Sensors, actuators, control theory
💰 4	İqtisadiyyatın Əsasları	Supply, demand, inflation, markets

⚙️ Training Details

Parameter	Value
🧱 Base Model	Gemma 3 (270M)
🎓 Fine-tuning Type	Instruction-tuning
💾 Precision	BF16
🔁 Epochs	3
🚀 Learning Rate	2e-5

🧠 Model Capabilities

✅ Understands and answers 🇦🇿 Azerbaijani instructions naturally
✅ Supports general academic and technical fields
✅ Performs simple reasoning and step-by-step explanations
✅ Compact enough for local deployment (mobile & web apps)

💬 Example Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Yusiko/LIA-AzInstruction-400")
tokenizer = AutoTokenizer.from_pretrained("Yusiko/LIA-AzInstruction-400")

prompt = "Qarabağ haqqında qısa məlumat ver."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=150)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Yusiko/AzInstruction-400

Base model

google/gemma-3-270m

Finetuned

(93)

this model

Yusiko
/

AzInstruction-400