Step 1: Install MoroAI (30 seconds)

Install the MoroAI core package with training extensions:

# Install with training support
pip install "moroai[train]"
  

Verify that your environment and GPU are ready:

moro doctor
  

Step 2: Initialize Your Project (30 seconds)

moro init my-first-model
cd my-first-model
  

This creates a reproducible workspace:

my-first-model/
├── moro.yaml          # Project configuration & recipe
├── data/
│   └── raw/           # Raw training data inputs
├── runs/              # Training checkpoints & telemetry
├── releases/          # Exported GGUF models & SBOMs
└── eval/              # Evaluation suites
  

Step 3: Add Your Data (1 minute)

Create a sample customer support dataset:

cat > data/raw/support_data.jsonl << 'EOF'
{"messages": [{"role": "user", "content": "How do I reset my password?"}, {"role": "assistant", "content": "Go to Settings > Security > Reset Password."}]}
{"messages": [{"role": "user", "content": "What is your refund policy?"}, {"role": "assistant", "content": "We offer a 30-day money-back guarantee on all purchases."}]}
{"messages": [{"role": "user", "content": "How do I contact support?"}, {"role": "assistant", "content": "You can reach our support team at support@example.com."}]}
EOF
  

Ensure moro.yaml references your raw data:

dataset:
  source: ./data/raw/support_data.jsonl
  format: jsonl
  

Step 4: Build the Dataset with MI Guard (30 seconds)

moro data build
  

✓ Loaded 3 samples
✓ After deduplication: 3 samples
✓ MI Guard: 1 rare domain landmark protected
✓ Split: train=2, val=1, eval=0
✓ Dataset compiled to data/compiled
  

Step 5: Train Your Model (2 minutes)

moro train
  

MoroAI detects your hardware, audits VRAM memory safety, and runs local fine-tuning:

✓ Hardware detected: NVIDIA GeForce RTX 4070 (12GB)
✓ Recipe generated: lora_r=16, lr=2e-4, batch=2, accum=8
✓ Training started: run_1234567890
✓ Step 50/100 | Loss: 1.42 | VRAM: 6.2GB
✓ Training completed in 45s (Zero OOM)
  

Step 6: Deploy to Ollama (30 seconds)

moro release export --run-id run_1234567890 --version v1.0.0
moro release deploy --target ollama --version v1.0.0
  

Step 7: Test Your Private Model (30 seconds)

# Test locally via Ollama CLI:
ollama run my-first-model:v1.0.0 "How do I reset my password?"

# Or via the OpenAI-compatible local REST API:
curl http://localhost:11434/api/generate -d '{
  "model": "my-first-model:v1.0.0",
  "prompt": "How do I reset my password?",
  "stream": false
}'
  

What Just Happened?