QUICKSTART.md

🚀 Quick Start Guide

Get DReamMachine running in 5 minutes!

Step 1: Install Dependencies

pip install -r requirements.txt

Step 2: Set Up HuggingFace Token

Get your token from: https://huggingface.co/settings/tokens

Option A: Environment Variable

export HF_TOKEN=your_token_here

Option B: .env File

cp .env.example .env
# Edit .env and add your token

Step 3: Verify Setup

python setup.py

You should see all checks pass ✓

Step 4: Run Your First Dream!

Option A: Gradio Web Interface (Recommended)

python app.py

Then open: http://localhost:7860

1. Go to "Single Dream Round" tab
2. Keep the default stage "init_1_25"
3. Click "🚀 Run Dream Round"
4. Wait 2-5 minutes
5. Review the amazing results!

Option B: Command Line

python run_cli.py --single

What Happens During a Dream Round?

[A.1] Setup → Load prompts and constraints
[A.2] Dream → 3 LLMs generate creative ideas
[A.3] Refine → Writer/Logger/Narrator create narrative
[A.4] Analyze → Deep Thinker evaluates feasibility
[A.5] Score → Curator assigns final grades
[A.6] Log → Save to local files + HF Dataset
[A.7] Decide → Advance idea or start new one?

Takes 2-5 minutes depending on API response times.

Understanding Your Results

Session Summary

Originality: 9/10       ← How novel is it?
Feasibility: 8/10       ← Can it be built?
Global Impact: 9/10     ← How many people benefit?
Narrative Coherence: 8/10 ← Is the story good?

Reforge Flag: ✓ Yes    ← Advances to next stage!

Reforge Flag = True means the idea was good enough to advance!

The Outputs

1. Narrative Pitch - Compelling story about the invention
2. Technical Components - What's needed to build it
3. Feasibility Report - Scientific evaluation
4. Curator Scorecard - Detailed scoring breakdown

Try Different Stages

Once you've run init_1_25 successfully:

# Try mid-life crisis stage
python run_cli.py --single --stage mid_26_50

# Try late-life adoption stage
python run_cli.py --single --stage late_51_75

# Try final legacy stage
python run_cli.py --single --stage final_76_100

Run Multiple Dreams (Batch Mode)

Via Web Interface

1. Go to "Batch Mode" tab
2. Set rounds: 5
3. Set interval: 30 seconds
4. Click "Run Batch Mode"
5. Get coffee ☕ (takes ~15 minutes)

Via CLI

# Run 5 rounds with 30-second intervals
python run_cli.py --batch 5 --interval 30

View Your Session History

Web Interface

1. Go to "Session History" tab
2. Click "🔄 Refresh History"
3. Browse all your past dreams!

File System

Check the logs/ directory:

ls logs/
# session_20250114_123045_0.json
# session_20250114_124512_1.json
# ...

Customize Configuration

Edit config.yaml to change:

• Models: Use different LLMs
• Constraints: Change what ideas must satisfy
• Thresholds: Adjust scoring requirements
• Prompts: Modify detail level

Common Issues

"HF_TOKEN not found" → Set the environment variable or add to .env file

"Rate limit exceeded" → Add more sleep time between rounds → Use --interval 60 for batch mode

"Model timeout" → Normal for large models, will retry automatically → Consider using smaller/faster models for testing

Next Steps

• Read the full README.md for advanced features
• Customize prompts in prompt_manager.py
• Deploy to HuggingFace Spaces for 24/7 dreaming
• Share your best discoveries!

---

Happy Dreaming! 🌟