DEPLOY.md

# 🚀 Vish AI - Deployment Guide

## 📦 Essential Files for Hugging Face Spaces

Upload these **3 files only**:

1. **`app.py`** (18KB) - Main application with Phi-3 integration
2. **`requirements.txt`** (515 bytes) - All dependencies
3. **`README.md`** (7.1KB) - Space description and info

**That's it!** Everything else is optional.

---

## 🎯 Quick Deploy Steps

### 1. Create Hugging Face Space

Go to: https://huggingface.co/new-space

Settings:
- **Owner**: Your username
- **Space name**: `vish-ai` (or your choice)
- **License**: MIT
- **SDK**: Gradio
- **Python version**: 3.10 or 3.11 ⚠️ (Required)
- **Hardware**: 
  - CPU Basic (FREE) - Works, but slower (3-6s responses)
  - T4 GPU ($0.60/hr) - Recommended (0.5-2s responses)

### 2. Upload Files

**Option A - Web Upload:**
1. Click "Files" tab in your new Space
2. Click "Add file" → "Upload files"
3. Upload: `app.py`, `requirements.txt`, `README.md`
4. Click "Commit to main"

**Option B - Git Clone:**
```bash
git clone https://huggingface.co/spaces/YOUR_USERNAME/vish-ai
cd vish-ai
cp /path/to/app.py .
cp /path/to/requirements.txt .
cp /path/to/README.md .
git add .
git commit -m "Deploy Vish AI with Phi-3"
git push
```

### 3. Wait for Build

⏱️ **First build takes 15-20 minutes:**
- Installing dependencies: ~3 min
- Downloading Phi-3 model (7GB): ~10-15 min
- Starting app: ~2 min

**Watch the logs** (click "Logs" tab) for:
```
📥 Loading Phi-3 Mini unified model...
   Model: microsoft/Phi-3-mini-4k-instruct
   This may take 5-15 minutes on first run (downloading ~7GB)...
   Loading tokenizer...
   ✅ Tokenizer loaded
   Loading model (this is the slow part)...
✅ Phi-3 Mini model loaded successfully!
🎉 Unified model ready for all tasks!
   Model parameters: 3,821,079,552

✅ All systems ready!

Running on local URL:  http://0.0.0.0:7860
```

### 4. Test Your Space

Once live, test all 3 features:

**Chat:**
- Input: "What is artificial intelligence?"
- Expected: Intelligent multi-paragraph response

**Summarize:**
- Input: Paste 100+ word article
- Expected: 2-3 sentence summary

**Sentiment:**
- Input: "I absolutely love this product!"
- Expected: "😊 POSITIVE"

---

## 🔐 Optional: Add Supabase Authentication

If you want user authentication and logging:

1. Go to Space Settings → "Variables and secrets"
2. Add these secrets:
   ```
   NEXT_PUBLIC_SUPABASE_URL = https://lyebtceryednzafhyunq.supabase.co
   NEXT_PUBLIC_SUPABASE_ANON_KEY = your_anon_key_here
   ```
3. Restart Space

Without Supabase: App works perfectly, just no user logging.

---

## 📊 Performance Expectations

| Hardware | Chat | Summarize | Sentiment |
|----------|------|-----------|-----------|
| CPU Basic (FREE) | 3-6s | 4-8s | 2-4s |
| T4 GPU Small | 0.5-1.5s | 1-2s | 0.3-0.8s |
| A10G GPU | 0.2-0.6s | 0.5-1s | 0.2-0.5s |

---

## 🔧 Troubleshooting

### ❌ Build fails with "Out of Memory"

**Fix**: Model is optimized for CPU. If still failing:
- Upgrade to T4 GPU (has more memory)
- Check logs for specific error

### ❌ "AI models not available" in app

**Fix**: 
- Wait for build to complete (full 20 minutes)
- Check logs for download progress
- Ensure Python 3.10 or 3.11 (not 3.12+)

### ❌ Slow responses (>10 seconds)

**Fix**:
- Normal on CPU Basic (AI is compute-intensive)
- Upgrade to T4 GPU for 5-10x speedup
- First response is slower (model warmup)

### ❌ Model not downloading

**Fix**:
- Check build logs for errors
- Ensure internet access (Spaces have it)
- Wait full 20 minutes before retrying

---

## 📁 Optional Files Explained

### `test_phi3_model.py` (8.3KB)
Test the model locally before deploying:
```bash
pip install -r requirements.txt
python test_phi3_model.py
```

### `fine_tune_phi3.py` (6.8KB)
Fine-tune Phi-3 on your custom data (advanced):
```bash
python fine_tune_phi3.py
```

### `supabase_setup.sql` (6KB)
SQL schema for Supabase database tables (if using auth).

---

## ✅ Success Checklist

- [ ] Space created on Hugging Face
- [ ] Python version is 3.10 or 3.11
- [ ] Files uploaded: `app.py`, `requirements.txt`, `README.md`
- [ ] Build completed without errors
- [ ] Logs show: "✅ Phi-3 Mini model loaded successfully!"
- [ ] Chat responds intelligently
- [ ] Summarizer condenses text
- [ ] Sentiment analyzer detects emotions
- [ ] Response times acceptable for your use case

---

## 🌐 Your Live Space

After deployment, share your app:
```
https://huggingface.co/spaces/YOUR_USERNAME/vish-ai
```

Example:
```
https://huggingface.co/spaces/vishwas896/vish-ai
```

---

## 💡 Pro Tips

1. **Start with CPU Basic** (free) for testing
2. **Monitor usage** - upgrade to GPU only if needed
3. **First response is slower** (model warmup) - this is normal
4. **Check logs regularly** during first build
5. **Test all features** before sharing publicly
6. **GPU pricing**: Only charged when Space is running
7. **Pause Space** when not in use to save costs (GPU only)

---

## � Need Help?

- **Build logs**: Check for detailed error messages
- **Hugging Face Docs**: https://huggingface.co/docs/hub/spaces
- **GitHub Issues**: Report problems in your repo

---

## 🎉 What You've Built

✨ **Powerful AI assistant** with:
- Microsoft Phi-3 Mini (3.8 billion parameters)
- Chat, Summarization, and Sentiment Analysis
- Clean, production-ready code
- Deployed on Hugging Face's infrastructure
- Optional user authentication with Supabase

**Total setup: Just 3 files, ~25KB. That's it!**

---

Built with ❤️ by Vishwas | Powered by Microsoft Phi-3 & Hugging Face