Novaai

CRANE AI Model

CRANE (Compressed Routing and Neural Embedding) hibrit yapay zeka sistemi.

Model Description

CRANE AI, 4 farklı özelleşmiş modülü birleştiren hibrit bir AI sistemidir:

• CodeModule: Kod yazma (DeepSeek-Coder 1.3B)
• ChatModule: Genel sohbet (Qwen2.5 1.5B)
• ReasonModule: Mantık yürütme (Phi-3 Mini)
• FastModule: Hızlı yanıtlar (TinyLlama 1.1B)

Quick Start

from crane_ai import CRANEAISystem
import asyncio

# Sistemi başlat
system = CRANEAISystem()
await system.initialize()

# Kullanım
response = await system.process_query("Python'da bir hesap makinesi yaz", {})
print(response["response"])

Installation

pip install -r requirements.txt
python setup.py install

Usage

API Sunucusu

python main.py

Programmatic Usage

from crane_ai import CRANEAISystem

system = CRANEAISystem()
await system.initialize()

result = await system.process_query("Your query here", {})

Fine-tuning

CRANE AI LoRA/QLoRA ile fine-tune edilebilir:

python training/fine_tune.py --module code_module --data your_data.jsonl

Model Architecture

Input Query → Router → Best Module → Response
     ↓
[CodeModule, ChatModule, ReasonModule, FastModule]
     ↓
Token Capsule Layer → Memory Management

Training Data

Base modeller açık kaynak dataseti ile eğitilmiş:

• Code: GitHub repositories
• Chat: Conversation datasets
• Reasoning: Logic puzzles
• Fast: Q&A pairs

Limitations

• GPU memory: ~4GB gerekli
• Response time: 1-5 saniye
• Context length: 4096 tokens max

License

MIT License - Commercial use allowed.

Citation

@misc{crane-ai-2024,
  title={CRANE AI: Hybrid Multi-Model System},
  author={Veteroner},
  year={2024},
  url={https://huggingface.co/veteroner/Novaai}
}