README.md

---
language: en
license: mit
tags:
- spiritual-ai
- brahma-kumaris
- murli
- distilgpt2
- ultra-lite
- peft
- lora
library_name: peft
base_model: distilgpt2
---

# 🕉️ Murli Assistant - DistilGPT-2 Ultra-Lite

An **ultra-lightweight** spiritual AI assistant trained on Brahma Kumaris murli content. Perfect for free Colab and low-resource environments!

## 🎯 Why This Model?

- **82M parameters** (30x smaller than Phi-2)
- **RAM: ~1-2 GB** (fits easily in free Colab)
- **Fast inference**: 0.5-1 second per response
- **No quantization needed**: Runs in full precision
- **Perfect for free tier**: No crashes, no OOM errors

## Model Details

- **Base Model**: DistilGPT-2 (82M parameters)
- **Fine-tuning**: LoRA (Low-Rank Adaptation)
- **Training Data**: 150 authentic murlis
- **Training Examples**: 153+
- **Max Length**: 256 tokens
- **LoRA Rank**: 4

## Usage

### Quick Start (Colab)

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load base model
tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
base_model = AutoModelForCausalLM.from_pretrained("distilgpt2")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "eswarankrishnamurthy/murli-assistant-distilgpt2-lite")

# Chat function
def chat(message):
    prompt = f"Q: {message}\nA:"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=150)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Try it
response = chat("Om Shanti")
print(response)
```

### Use in Production

See the full Colab notebook: `murli-distilgpt2-colab.ipynb`

## Comparison with Other Models

| Model | Parameters | RAM | Inference | Colab Free |
|-------|------------|-----|-----------|------------|
| **DistilGPT-2 (This)** | 82M | ~1-2 GB | 0.5-1s | ✅ Perfect |
| Phi-2 | 2.7B | ~10 GB | 1-3s | ❌ Crashes |
| Phi-2 (4-bit) | 2.7B | ~3-4 GB | 1-3s | ⚠️ Tight fit |

## Advantages

✅ **Ultra-Lightweight**: 30x smaller than Phi-2
✅ **Low RAM**: Only 1-2 GB needed
✅ **Fast Training**: 5-10 minutes
✅ **Fast Inference**: Sub-second responses
✅ **Free Colab**: Perfect fit, no crashes
✅ **Easy Deployment**: Simple integration
✅ **Good Quality**: Excellent for basic Q&A

## Training Details

[
  "30x smaller than Phi-2",
  "Fits in free Colab RAM easily",
  "Fast training (5-10 min)",
  "Fast inference",
  "Good for basic Q&A"
]

## Example Responses

**Q:** Om Shanti
**A:** Om Shanti, sweet child! 🙏 I'm your Murli Helper. How can I guide you today?

**Q:** What is soul consciousness?
**A:** Soul consciousness is experiencing yourself as an eternal, pure soul with peace, love, and purity. Om Shanti 🙏

**Q:** Who is Baba?
**A:** Baba is the Supreme Soul, the Ocean of Knowledge who teaches Raja Yoga through Brahma. Om Shanti 🙏

## Limitations

- Shorter context (256 tokens vs Phi-2's 512)
- Simpler responses compared to larger models
- Best for focused Q&A, not long essays
- Limited reasoning compared to billion-parameter models

## License

MIT License - Free to use and modify

## Citation

```bibtex
@misc{murli-distilgpt2-lite,
  author = {eswarankrishnamurthy},
  title = {Murli Assistant - DistilGPT-2 Ultra-Lite},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/eswarankrishnamurthy/murli-assistant-distilgpt2-lite}
}
```

## Acknowledgments

- Brahma Kumaris World Spiritual University for murli teachings
- HuggingFace for model hosting
- DistilGPT-2 team for the base model

---

**Om Shanti! 🙏**