For AI Researchers
Find licensed training data, pay per epoch, and receive your fine-tuned model — without managing infrastructure.
For AI Researchers
LICEN abstracts away everything you don't want to deal with: legal licensing, data procurement, storage downloads, GPU setup, and key management.
You find data you want. You review the publisher's usage policy. You pay for the compute you need. You get a fine-tuned model.
Before You Start
You'll need:
- A Web3 wallet (LICEN uses Privy — MetaMask or email login both work)
- 0G tokens for escrow (the payment locked until your training job completes)
Get testnet tokens from the 0G Faucet →
Finding a Dataset
Go to the Marketplace. Every listed dataset is a live on-chain entry — the prices and policies you see are enforced by smart contract, not manually managed.
Each dataset card shows:
- Title and description — what the dataset contains and what it's good for
- Access policy — the usage boundary the publisher has defined for this dataset
- Allowed purposes — what the publisher has permitted (e.g. medical research only, no commercial use)
- Price per epoch — the cost in 0G tokens per training pass
- Max epochs per run — the cap the publisher has set for a single session
- Max runs per requester — whether repeat access is limited at the wallet level
- Session TTL — how long your access grant remains valid once approved
Filter by purpose, price, or availability. Click any dataset to see its full policy. Purpose is only one part of that policy; the full policy also determines the allowed volume, duration, and repeatability of access.
Requesting Access
When you find a dataset you want:
- Click Request Access
- Set the number of epochs you want to train for
- Review the cost breakdown:
Total escrow = price per epoch × requested epochs- Click Submit and approve the transaction in your wallet
That's it. The smart contract locks your escrow, validates your request against the publisher's full usage policy, and triggers the training pipeline automatically.
What Happens Next (automatically)
After your transaction confirms, LICEN's Orchestrator takes over:
| Step | What's happening |
|---|---|
| Requested | Your escrow is locked, the Orchestrator is picking up the job |
| Running | The demo is simulating the 0G Compute training lifecycle; production will run this inside an attested 0G-compatible confidential node |
| Completed | Training finished, your model is on 0G Storage, settlement is processing |
| Failed | Something went wrong — you'll be automatically refunded |
You don't need to do anything. Track progress in your Sessions tab.
Getting Your Model
When the job reaches Completed:
- Your session view shows a Result Hash — the 0G Storage root of your trained model
- The model is a LoRA adapter for the base model that was trained on (e.g. Qwen2.5-0.5B-Instruct)
- Download it and load it on top of the base model for inference
from peft import PeftModel
from transformers import AutoModelForCausalLM
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
model = PeftModel.from_pretrained(base, "./your-lora-adapter")Escrow & Refunds
You pay upfront based on the epochs you requested. Settlement is based on the epochs actually completed:
- If training ran the full requested epochs → you pay the full escrow
- If training stopped early (compute failure, early convergence) → the unused portion is automatically refunded to your wallet
The DataPolicy contract handles this settlement — no support ticket required.