Question 1

How do I submit my model training code to G.O.D tournaments?

Accepted Answer

Register a hotkey on the G.O.D subnet (netuid 56), post your miner IP to the metagraph, then implement the `/training_repo/{task_type}` endpoint in your miner to return a GitHub repository URL, full 40-character commit SHA, and optional read-only token. The validator queries your endpoint during registration and validates the commit contains required Dockerfiles (one per task type), LICENSE, and NOTICE files matching the subnet's standard.

Question 2

What compute resources do miners need to provide?

Accepted Answer

Miners do not provide compute. You only need a running miner endpoint (on port 7999) so validators can request your repository details. Validators manage all GPU infrastructure, cloning your code, building Docker images, running training, and uploading results. You pay a per-tournament entry fee (0.4–0.7 TAO depending on type) deducted from your coldkey balance.

Question 3

How are tournament winners determined and paid?

Accepted Answer

Tournaments progress through group, knockout, and final boss rounds. Text and image tasks score on sample-wise loss comparison in knockout rounds, with challengers requiring 55% win rate on decided samples plus mean-loss improvement to advance. Boss rounds add stricter criteria: 55% win rate on held-out samples plus scaled mean-loss gap. Winners earn weight (1.0–3.0x base per-type weight depending on rank and performance) which translates to weekly emissions. Top three ranks are paid via exponential decay (roughly 76/19/5 split).

Question 4

Can I use private repositories or request additional datasets?

Accepted Answer

Yes. Return a GitHub fine-grained personal access token with read-only Contents access in your repository response, and the validator will clone private repos. You can also request up to two Hugging Face datasets from a whitelist; validators pre-download them into `/cache/miner_datasets/` and pass the directory path via environment variables during training.

Question 5

What happens to my code if my submission wins?

Accepted Answer

Winning repositories are re-uploaded to the public Gradients organization (github.com/gradients-opensource) with your authorship retained. Only submit code you are prepared to open-source if it wins, and ensure your submission includes a valid LICENSE and NOTICE matching the subnet's standard to pass validation.

Question 6

How does the subnet prevent duplicate or obfuscated submissions?

Accepted Answer

Submissions are de-duplicated across three tiers: identical commit SHA, source after stripping whitespace/comments (T1), and Claude pairwise code review (T2). Cosmetic disguises (renaming, reordering, noise hyperparameters, dead code) are treated as duplicates and eliminated. Repositories are also scanned for obfuscation (compiled/minified code, hidden datasets, pretrained models) at validation time and rejected if found.

Gradients

Gradients

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup

Participate

Participate

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup