Developer Tutorials

How to Deploy Your Own AI Model as a Miner on ORI

ORI allows you to deploy your custom AI models (like your TRM Oracle) as miners. Your model runs on confidential Targon infrastructure, but you only interact with ORI’s clean dashboard and API. No Targon commands or SDK are visible to you.

Prerequisites

• ORI Miner account (enabled)
• Your model files ready (e.g. miner.py, trm_100_oracle.py, hybrid_heads.py, ETH_trm_oracle.pt, ETH_oracles.json, etc.)
• Basic Python knowledge

Step 1: Prepare Your Miner Package

Create a clean folder with only the files needed for inference:

eth_miner_v1/
├── core
│   ├── hybrid_heads.py
│   ├── __init__.py
│   └── trm_100_oracle.py
├── ETH_01_correlation_hunter.py (only for reference)
├── ETH_02_dataset_builder.py (only for reference)
├── ETH_03_signal_generator_training.py (only for reference)
├── ETH_04b_xgboost_baseline.py (only for reference)
├── ETH_04_evaluator.py (only for reference)
├── ETH_18_signal_generator_daemon.py (only for reference)
├── ETH_19_evaluation_truth_engine.py (only for reference)
├── ETH_ml_dataset.csv (only for reference)
├── ETH_oracles.json (model specs: e.g.:
                        {
                            "target": "ETH-USDC",
                            "optimal_horizon": 66,
                            "oracles": {
                                "BTC-USDC": 66,
                                "DOGE-USDC": 61,
                                "LINK-USDC": 149,
                                "SPY": 109
                            }
                        }
                        )
├── ETH_raw_data.csv (only for reference)
├── ETH_trm_oracle.pt (model weights)
├── miner.py
├── README.md
└── requirements.txt

Important modification to miner.py
Replace the dual-broadcast section with a clean call to ORI (remove the local 127.0.0.1:5050 hardcoding if you want it fully portable):

# Inside forward() method, replace the _push_to_platform block with:

try:
    import threading
    def _push_to_platform():
        try:
            payload = {
                "model_id": self.wallet.hotkey.ss58_address,
                "asset": self.asset,
                "spot_price": float(current_spot),
                "ptdv": ptdv_signal,
                "prediction_timestamp": synapse.prediction_timestamp,
                "processing_time": time.time() - start_time
            }
            # This will be replaced by ORI during deployment with the correct endpoint
            import requests
            requests.post("https://api.ori.app/api/v1/submit_ptdv", 
                         json=payload, timeout=8)
        except Exception as e:
            pass  # Silent fail - do not crash the miner
    threading.Thread(target=_push_to_platform, daemon=True).start()
except Exception:
    pass

Step 2: Go to ORI Miner Dashboard

Log in → Miner → Deploy New Model

Step 3: Use the Deployment Wizard

Choose “Advanced Custom Miner (Bittensor-style)”

• Upload Package: Zip your my-miner/ folder and upload it
• Entry Point: miner.py (the file containing the ETHMiner class)
• Main Class: ETHMiner
• Hardware Tier: GPU (recommended for TRM Oracle) or CPU Small
• Signal Endpoint: ORI will automatically expose /generate_signal (or you can keep using the internal Bittensor axon style — ORI handles the wrapper)

ORI will:

• Automatically install dependencies from requirements.txt
• Bundle your model weights (ETH_trm_oracle.pt)
• Wrap your miner so it can receive challenges and submit PTDV signals via ORI’s API
• Deploy it as a confidential, auto-scaling serverless function on Targon

Step 4: Test the Deployment

After deployment:

• Use the Test Challenge button in ORI
• Send a sample TRMChallenge payload
• Check that you receive a correctly formatted PTDV response and that it appears in your platform’s signal feed

Step 5: Register for Competition

Click “Register Model for Live Competition”.
Your miner is now eligible to receive real challenges and earn based on signal quality.

Monitoring

In My Models you can see:

• Status & uptime
• Number of challenges processed
• Average processing time
• Tokens/compute usage (for billing transparency)