🧠 Creaton AI — Fine-Tuned Phi-2 Model for Ollama

Creaton AI is a locally runnable, instruction-tuned variant of Microsoft’s Phi-2 language model. Fine-tuned using the LoRA method and converted to gguf format for compatibility with Ollama, it supports fast and efficient inference on consumer hardware.

---

📌 Model Summary

• Base Model: microsoft/phi-2
• Fine-tuned With: LoRA (Low-Rank Adaptation)
• Quantization: 4-bit GGUF (creaton-ai.gguf)
• Format: Instruction → Output (Instruct-style)
• Inference Engine: Ollama

---

🚀 How to Use

1. ✅ Prerequisites

• Install Ollama (Windows, Mac, Linux)
• Download the creaton-ai.gguf and place it next to the Modelfile

---

2. 📦 Build the Model

From the folder containing Modelfile and creaton-ai.gguf:

ollama create creaton-ai -f Modelfile
````

> If successful: `✓ Created model creaton-ai`

---

### 3. 💬 Run the Model

```bash
ollama run creaton-ai

Example prompt:

Instruct: Explain the purpose of the Indian Constitution.
Output:

---

🔌 API Access via REST

Start Ollama’s server:

ollama serve

Then query via curl:

curl http://localhost:11434/api/generate -d '{
  "model": "creaton-ai",
  "prompt": "Instruct: Describe Thirukkural 133.\nOutput:",
  "stream": false
}'

---

📚 Citations & References

1. Microsoft Phi-2 Yao et al. (2023). Phi-2: The surprising power of small language models → https://huggingface.co/microsoft/phi-2

2. LoRA: Low-Rank Adaptation of LLMs Hu et al. (2021). arXiv:2106.09685

3. Transformers Library Wolf et al. (2020). https://github.com/huggingface/transformers

4. llama.cpp / GGUF Format Georgi Gerganov (2023). https://github.com/ggerganov/llama.cpp

5. Ollama https://ollama.com

---

⚠️ Disclaimer

This model is provided for educational and research purposes only. Redistribution must comply with the license of the base model and the Ollama runtime.

---

🙌 Acknowledgments

• Microsoft for Phi-2
• Hugging Face and 🤗 Transformers
• llama.cpp & GGUF contributors
• Ollama Team

---